Upload folder using huggingface_hub

.gitattributes

@@ -60,3 +60,77 @@ lyric-generator-ja/frontend/public/font/materialdesignicons-webfont.ttf filter=l
 lyric-generator-ja/frontend/public/font/materialdesignicons-webfont.woff filter=lfs diff=lfs merge=lfs -text
 lyric-generator-ja/frontend/public/font/materialdesignicons-webfont.woff2 filter=lfs diff=lfs merge=lfs -text
 lyric-generator-ja/frontend/public/font/materialdesignicons-webfont_ie.eot filter=lfs diff=lfs merge=lfs -text
+BanG-Dream-VITS/BANNER.PNG filter=lfs diff=lfs merge=lfs -text
+BanG-Dream-VITS/ave-mujica-avemuji.gif filter=lfs diff=lfs merge=lfs -text
+BanG-Dream-VITS/image/あこ.png filter=lfs diff=lfs merge=lfs -text
+BanG-Dream-VITS/image/こころ.png filter=lfs diff=lfs merge=lfs -text
+BanG-Dream-VITS/image/そよ.png filter=lfs diff=lfs merge=lfs -text
+BanG-Dream-VITS/image/たえ.png filter=lfs diff=lfs merge=lfs -text
+BanG-Dream-VITS/image/つくし.png filter=lfs diff=lfs merge=lfs -text
+BanG-Dream-VITS/image/つぐみ.png filter=lfs diff=lfs merge=lfs -text
+BanG-Dream-VITS/image/にゃむ.png filter=lfs diff=lfs merge=lfs -text
+BanG-Dream-VITS/image/はぐみ.png filter=lfs diff=lfs merge=lfs -text
+BanG-Dream-VITS/image/ひまり.png filter=lfs diff=lfs merge=lfs -text
+BanG-Dream-VITS/image/ましろ.png filter=lfs diff=lfs merge=lfs -text
+BanG-Dream-VITS/image/りみ.png filter=lfs diff=lfs merge=lfs -text
+BanG-Dream-VITS/image/イヴ.png filter=lfs diff=lfs merge=lfs -text
+BanG-Dream-VITS/image/チュチュ.png filter=lfs diff=lfs merge=lfs -text
+BanG-Dream-VITS/image/パレオ.png filter=lfs diff=lfs merge=lfs -text
+BanG-Dream-VITS/image/モカ.png filter=lfs diff=lfs merge=lfs -text
+BanG-Dream-VITS/image/リサ.png filter=lfs diff=lfs merge=lfs -text
+BanG-Dream-VITS/image/レイヤ.png filter=lfs diff=lfs merge=lfs -text
+BanG-Dream-VITS/image/ロック.png filter=lfs diff=lfs merge=lfs -text
+BanG-Dream-VITS/image/一愛.png filter=lfs diff=lfs merge=lfs -text
+BanG-Dream-VITS/image/七深.png filter=lfs diff=lfs merge=lfs -text
+BanG-Dream-VITS/image/光.png filter=lfs diff=lfs merge=lfs -text
+BanG-Dream-VITS/image/克洛迪娜.png filter=lfs diff=lfs merge=lfs -text
+BanG-Dream-VITS/image/初華.png filter=lfs diff=lfs merge=lfs -text
+BanG-Dream-VITS/image/千聖.png filter=lfs diff=lfs merge=lfs -text
+BanG-Dream-VITS/image/友希那.png filter=lfs diff=lfs merge=lfs -text
+BanG-Dream-VITS/image/司.png filter=lfs diff=lfs merge=lfs -text
+BanG-Dream-VITS/image/壘.png filter=lfs diff=lfs merge=lfs -text
+BanG-Dream-VITS/image/奈奈.png filter=lfs diff=lfs merge=lfs -text
+BanG-Dream-VITS/image/彩.png filter=lfs diff=lfs merge=lfs -text
+BanG-Dream-VITS/image/愛音.png filter=lfs diff=lfs merge=lfs -text
+BanG-Dream-VITS/image/文.png filter=lfs diff=lfs merge=lfs -text
+BanG-Dream-VITS/image/日菜.png filter=lfs diff=lfs merge=lfs -text
+BanG-Dream-VITS/image/晶.png filter=lfs diff=lfs merge=lfs -text
+BanG-Dream-VITS/image/有咲.png filter=lfs diff=lfs merge=lfs -text
+BanG-Dream-VITS/image/未知留.png filter=lfs diff=lfs merge=lfs -text
+BanG-Dream-VITS/image/栞.png filter=lfs diff=lfs merge=lfs -text
+BanG-Dream-VITS/image/楽奈.png filter=lfs diff=lfs merge=lfs -text
+BanG-Dream-VITS/image/沙綾.png filter=lfs diff=lfs merge=lfs -text
+BanG-Dream-VITS/image/海鈴.png filter=lfs diff=lfs merge=lfs -text
+BanG-Dream-VITS/image/燈.png filter=lfs diff=lfs merge=lfs -text
+BanG-Dream-VITS/image/燐子.png filter=lfs diff=lfs merge=lfs -text
+BanG-Dream-VITS/image/珠緒.png filter=lfs diff=lfs merge=lfs -text
+BanG-Dream-VITS/image/真晝.png filter=lfs diff=lfs merge=lfs -text
+BanG-Dream-VITS/image/真矢.png filter=lfs diff=lfs merge=lfs -text
+BanG-Dream-VITS/image/睦.png filter=lfs diff=lfs merge=lfs -text
+BanG-Dream-VITS/image/祥子.png filter=lfs diff=lfs merge=lfs -text
+BanG-Dream-VITS/image/立希.png filter=lfs diff=lfs merge=lfs -text
+BanG-Dream-VITS/image/純那.png filter=lfs diff=lfs merge=lfs -text
+BanG-Dream-VITS/image/紗夜.png filter=lfs diff=lfs merge=lfs -text
+BanG-Dream-VITS/image/美咲.png filter=lfs diff=lfs merge=lfs -text
+BanG-Dream-VITS/image/美帆.png filter=lfs diff=lfs merge=lfs -text
+BanG-Dream-VITS/image/美空.png filter=lfs diff=lfs merge=lfs -text
+BanG-Dream-VITS/image/艾露.png filter=lfs diff=lfs merge=lfs -text
+BanG-Dream-VITS/image/艾露露.png filter=lfs diff=lfs merge=lfs -text
+BanG-Dream-VITS/image/花音.png filter=lfs diff=lfs merge=lfs -text
+BanG-Dream-VITS/image/菈樂菲.png filter=lfs diff=lfs merge=lfs -text
+BanG-Dream-VITS/image/華戀.png filter=lfs diff=lfs merge=lfs -text
+BanG-Dream-VITS/image/薫.png filter=lfs diff=lfs merge=lfs -text
+BanG-Dream-VITS/image/蘭.png filter=lfs diff=lfs merge=lfs -text
+BanG-Dream-VITS/image/透子.png filter=lfs diff=lfs merge=lfs -text
+BanG-Dream-VITS/image/雙葉.png filter=lfs diff=lfs merge=lfs -text
+BanG-Dream-VITS/image/靜羽.png filter=lfs diff=lfs merge=lfs -text
+BanG-Dream-VITS/image/香子.png filter=lfs diff=lfs merge=lfs -text
+BanG-Dream-VITS/image/香澄.png filter=lfs diff=lfs merge=lfs -text
+BanG-Dream-VITS/image/麻弥.png filter=lfs diff=lfs merge=lfs -text
+BanG-Dream-VITS/img/yuyu.png filter=lfs diff=lfs merge=lfs -text
+BanG-Dream-VITS/img/参数说明.png filter=lfs diff=lfs merge=lfs -text
+BanG-Dream-VITS/img/宵宫.png filter=lfs diff=lfs merge=lfs -text
+BanG-Dream-VITS/img/微信图片_20231010105112.png filter=lfs diff=lfs merge=lfs -text
+BanG-Dream-VITS/img/神里绫华.png filter=lfs diff=lfs merge=lfs -text
+BanG-Dream-VITS/img/纳西妲.png filter=lfs diff=lfs merge=lfs -text
+BanG-Dream-VITS/temp.wav filter=lfs diff=lfs merge=lfs -text

BanG-Dream-VITS/.gitattributes

@@ -0,0 +1,110 @@
+*.7z filter=lfs diff=lfs merge=lfs -text
+*.arrow filter=lfs diff=lfs merge=lfs -text
+*.bin filter=lfs diff=lfs merge=lfs -text
+*.bz2 filter=lfs diff=lfs merge=lfs -text
+*.ckpt filter=lfs diff=lfs merge=lfs -text
+*.ftz filter=lfs diff=lfs merge=lfs -text
+*.gz filter=lfs diff=lfs merge=lfs -text
+*.h5 filter=lfs diff=lfs merge=lfs -text
+*.joblib filter=lfs diff=lfs merge=lfs -text
+*.lfs.* filter=lfs diff=lfs merge=lfs -text
+*.mlmodel filter=lfs diff=lfs merge=lfs -text
+*.model filter=lfs diff=lfs merge=lfs -text
+*.msgpack filter=lfs diff=lfs merge=lfs -text
+*.npy filter=lfs diff=lfs merge=lfs -text
+*.npz filter=lfs diff=lfs merge=lfs -text
+*.onnx filter=lfs diff=lfs merge=lfs -text
+*.ot filter=lfs diff=lfs merge=lfs -text
+*.parquet filter=lfs diff=lfs merge=lfs -text
+*.pb filter=lfs diff=lfs merge=lfs -text
+*.pickle filter=lfs diff=lfs merge=lfs -text
+*.pkl filter=lfs diff=lfs merge=lfs -text
+*.pt filter=lfs diff=lfs merge=lfs -text
+*.pth filter=lfs diff=lfs merge=lfs -text
+*.rar filter=lfs diff=lfs merge=lfs -text
+*.safetensors filter=lfs diff=lfs merge=lfs -text
+saved_model/**/* filter=lfs diff=lfs merge=lfs -text
+*.tar.* filter=lfs diff=lfs merge=lfs -text
+*.tar filter=lfs diff=lfs merge=lfs -text
+*.tflite filter=lfs diff=lfs merge=lfs -text
+*.tgz filter=lfs diff=lfs merge=lfs -text
+*.wasm filter=lfs diff=lfs merge=lfs -text
+*.xz filter=lfs diff=lfs merge=lfs -text
+*.zip filter=lfs diff=lfs merge=lfs -text
+*.zst filter=lfs diff=lfs merge=lfs -text
+*tfevents* filter=lfs diff=lfs merge=lfs -text
+image/あこ.png filter=lfs diff=lfs merge=lfs -text
+image/こころ.png filter=lfs diff=lfs merge=lfs -text
+image/そよ.png filter=lfs diff=lfs merge=lfs -text
+image/たえ.png filter=lfs diff=lfs merge=lfs -text
+image/つくし.png filter=lfs diff=lfs merge=lfs -text
+image/つぐみ.png filter=lfs diff=lfs merge=lfs -text
+image/にゃむ.png filter=lfs diff=lfs merge=lfs -text
+image/はぐみ.png filter=lfs diff=lfs merge=lfs -text
+image/ひまり.png filter=lfs diff=lfs merge=lfs -text
+image/ましろ.png filter=lfs diff=lfs merge=lfs -text
+image/りみ.png filter=lfs diff=lfs merge=lfs -text
+image/イヴ.png filter=lfs diff=lfs merge=lfs -text
+image/チュチュ.png filter=lfs diff=lfs merge=lfs -text
+image/パレオ.png filter=lfs diff=lfs merge=lfs -text
+image/モカ.png filter=lfs diff=lfs merge=lfs -text
+image/リサ.png filter=lfs diff=lfs merge=lfs -text
+image/レイヤ.png filter=lfs diff=lfs merge=lfs -text
+image/ロック.png filter=lfs diff=lfs merge=lfs -text
+image/一愛.png filter=lfs diff=lfs merge=lfs -text
+image/七深.png filter=lfs diff=lfs merge=lfs -text
+image/光.png filter=lfs diff=lfs merge=lfs -text
+image/克洛迪娜.png filter=lfs diff=lfs merge=lfs -text
+image/初華.png filter=lfs diff=lfs merge=lfs -text
+image/千聖.png filter=lfs diff=lfs merge=lfs -text
+image/友希那.png filter=lfs diff=lfs merge=lfs -text
+image/司.png filter=lfs diff=lfs merge=lfs -text
+image/壘.png filter=lfs diff=lfs merge=lfs -text
+image/奈奈.png filter=lfs diff=lfs merge=lfs -text
+image/彩.png filter=lfs diff=lfs merge=lfs -text
+image/愛音.png filter=lfs diff=lfs merge=lfs -text
+image/文.png filter=lfs diff=lfs merge=lfs -text
+image/日菜.png filter=lfs diff=lfs merge=lfs -text
+image/晶.png filter=lfs diff=lfs merge=lfs -text
+image/有咲.png filter=lfs diff=lfs merge=lfs -text
+image/未知留.png filter=lfs diff=lfs merge=lfs -text
+image/栞.png filter=lfs diff=lfs merge=lfs -text
+image/楽奈.png filter=lfs diff=lfs merge=lfs -text
+image/沙綾.png filter=lfs diff=lfs merge=lfs -text
+image/海鈴.png filter=lfs diff=lfs merge=lfs -text
+image/燈.png filter=lfs diff=lfs merge=lfs -text
+image/燐子.png filter=lfs diff=lfs merge=lfs -text
+image/珠緒.png filter=lfs diff=lfs merge=lfs -text
+image/真晝.png filter=lfs diff=lfs merge=lfs -text
+image/真矢.png filter=lfs diff=lfs merge=lfs -text
+image/睦.png filter=lfs diff=lfs merge=lfs -text
+image/祥子.png filter=lfs diff=lfs merge=lfs -text
+image/立希.png filter=lfs diff=lfs merge=lfs -text
+image/純那.png filter=lfs diff=lfs merge=lfs -text
+image/紗夜.png filter=lfs diff=lfs merge=lfs -text
+image/美咲.png filter=lfs diff=lfs merge=lfs -text
+image/美帆.png filter=lfs diff=lfs merge=lfs -text
+image/美空.png filter=lfs diff=lfs merge=lfs -text
+image/艾露.png filter=lfs diff=lfs merge=lfs -text
+image/艾露露.png filter=lfs diff=lfs merge=lfs -text
+image/花音.png filter=lfs diff=lfs merge=lfs -text
+image/菈樂菲.png filter=lfs diff=lfs merge=lfs -text
+image/華戀.png filter=lfs diff=lfs merge=lfs -text
+image/薫.png filter=lfs diff=lfs merge=lfs -text
+image/蘭.png filter=lfs diff=lfs merge=lfs -text
+image/透子.png filter=lfs diff=lfs merge=lfs -text
+image/雙葉.png filter=lfs diff=lfs merge=lfs -text
+image/靜羽.png filter=lfs diff=lfs merge=lfs -text
+image/香子.png filter=lfs diff=lfs merge=lfs -text
+image/香澄.png filter=lfs diff=lfs merge=lfs -text
+image/麻弥.png filter=lfs diff=lfs merge=lfs -text
+img/yuyu.png filter=lfs diff=lfs merge=lfs -text
+img/参数说明.png filter=lfs diff=lfs merge=lfs -text
+img/宵宫.png filter=lfs diff=lfs merge=lfs -text
+img/微信图片_20231010105112.png filter=lfs diff=lfs merge=lfs -text
+img/神里绫华.png filter=lfs diff=lfs merge=lfs -text
+img/纳西妲.png filter=lfs diff=lfs merge=lfs -text
+temp.wav filter=lfs diff=lfs merge=lfs -text
+ave-mujica-avemuji.gif filter=lfs diff=lfs merge=lfs -text
+d16ca6d1c911d68e1c6a998d893356f4.jpg filter=lfs diff=lfs merge=lfs -text
+BANNER.PNG filter=lfs diff=lfs merge=lfs -text

BanG-Dream-VITS/.pre-commit-config.yaml

@@ -0,0 +1,25 @@
+repos:
+  - repo: https://github.com/pre-commit/pre-commit-hooks
+    rev: v4.5.0
+    hooks:
+      - id: check-yaml
+      - id: end-of-file-fixer
+      - id: trailing-whitespace
+
+  - repo: https://github.com/astral-sh/ruff-pre-commit
+    rev: v0.1.7
+    hooks:
+      - id: ruff
+        args: [ --fix ]
+
+  - repo: https://github.com/psf/black
+    rev: 23.11.0
+    hooks:
+      - id: black
+
+  - repo: https://github.com/codespell-project/codespell
+    rev: v2.2.6
+    hooks:
+      - id: codespell
+        files: ^.*\.(py|md|rst|yml)$
+        args: [-L=fro]

BanG-Dream-VITS/LICENSE

@@ -0,0 +1,661 @@
+                    GNU AFFERO GENERAL PUBLIC LICENSE
+                       Version 3, 19 November 2007
+
+ Copyright (C) 2007 Free Software Foundation, Inc. <https://fsf.org/>
+ Everyone is permitted to copy and distribute verbatim copies
+ of this license document, but changing it is not allowed.
+
+                            Preamble
+
+  The GNU Affero General Public License is a free, copyleft license for
+software and other kinds of works, specifically designed to ensure
+cooperation with the community in the case of network server software.
+
+  The licenses for most software and other practical works are designed
+to take away your freedom to share and change the works.  By contrast,
+our General Public Licenses are intended to guarantee your freedom to
+share and change all versions of a program--to make sure it remains free
+software for all its users.
+
+  When we speak of free software, we are referring to freedom, not
+price.  Our General Public Licenses are designed to make sure that you
+have the freedom to distribute copies of free software (and charge for
+them if you wish), that you receive source code or can get it if you
+want it, that you can change the software or use pieces of it in new
+free programs, and that you know you can do these things.
+
+  Developers that use our General Public Licenses protect your rights
+with two steps: (1) assert copyright on the software, and (2) offer
+you this License which gives you legal permission to copy, distribute
+and/or modify the software.
+
+  A secondary benefit of defending all users' freedom is that
+improvements made in alternate versions of the program, if they
+receive widespread use, become available for other developers to
+incorporate.  Many developers of free software are heartened and
+encouraged by the resulting cooperation.  However, in the case of
+software used on network servers, this result may fail to come about.
+The GNU General Public License permits making a modified version and
+letting the public access it on a server without ever releasing its
+source code to the public.
+
+  The GNU Affero General Public License is designed specifically to
+ensure that, in such cases, the modified source code becomes available
+to the community.  It requires the operator of a network server to
+provide the source code of the modified version running there to the
+users of that server.  Therefore, public use of a modified version, on
+a publicly accessible server, gives the public access to the source
+code of the modified version.
+
+  An older license, called the Affero General Public License and
+published by Affero, was designed to accomplish similar goals.  This is
+a different license, not a version of the Affero GPL, but Affero has
+released a new version of the Affero GPL which permits relicensing under
+this license.
+
+  The precise terms and conditions for copying, distribution and
+modification follow.
+
+                       TERMS AND CONDITIONS
+
+  0. Definitions.
+
+  "This License" refers to version 3 of the GNU Affero General Public License.
+
+  "Copyright" also means copyright-like laws that apply to other kinds of
+works, such as semiconductor masks.
+
+  "The Program" refers to any copyrightable work licensed under this
+License.  Each licensee is addressed as "you".  "Licensees" and
+"recipients" may be individuals or organizations.
+
+  To "modify" a work means to copy from or adapt all or part of the work
+in a fashion requiring copyright permission, other than the making of an
+exact copy.  The resulting work is called a "modified version" of the
+earlier work or a work "based on" the earlier work.
+
+  A "covered work" means either the unmodified Program or a work based
+on the Program.
+
+  To "propagate" a work means to do anything with it that, without
+permission, would make you directly or secondarily liable for
+infringement under applicable copyright law, except executing it on a
+computer or modifying a private copy.  Propagation includes copying,
+distribution (with or without modification), making available to the
+public, and in some countries other activities as well.
+
+  To "convey" a work means any kind of propagation that enables other
+parties to make or receive copies.  Mere interaction with a user through
+a computer network, with no transfer of a copy, is not conveying.
+
+  An interactive user interface displays "Appropriate Legal Notices"
+to the extent that it includes a convenient and prominently visible
+feature that (1) displays an appropriate copyright notice, and (2)
+tells the user that there is no warranty for the work (except to the
+extent that warranties are provided), that licensees may convey the
+work under this License, and how to view a copy of this License.  If
+the interface presents a list of user commands or options, such as a
+menu, a prominent item in the list meets this criterion.
+
+  1. Source Code.
+
+  The "source code" for a work means the preferred form of the work
+for making modifications to it.  "Object code" means any non-source
+form of a work.
+
+  A "Standard Interface" means an interface that either is an official
+standard defined by a recognized standards body, or, in the case of
+interfaces specified for a particular programming language, one that
+is widely used among developers working in that language.
+
+  The "System Libraries" of an executable work include anything, other
+than the work as a whole, that (a) is included in the normal form of
+packaging a Major Component, but which is not part of that Major
+Component, and (b) serves only to enable use of the work with that
+Major Component, or to implement a Standard Interface for which an
+implementation is available to the public in source code form.  A
+"Major Component", in this context, means a major essential component
+(kernel, window system, and so on) of the specific operating system
+(if any) on which the executable work runs, or a compiler used to
+produce the work, or an object code interpreter used to run it.
+
+  The "Corresponding Source" for a work in object code form means all
+the source code needed to generate, install, and (for an executable
+work) run the object code and to modify the work, including scripts to
+control those activities.  However, it does not include the work's
+System Libraries, or general-purpose tools or generally available free
+programs which are used unmodified in performing those activities but
+which are not part of the work.  For example, Corresponding Source
+includes interface definition files associated with source files for
+the work, and the source code for shared libraries and dynamically
+linked subprograms that the work is specifically designed to require,
+such as by intimate data communication or control flow between those
+subprograms and other parts of the work.
+
+  The Corresponding Source need not include anything that users
+can regenerate automatically from other parts of the Corresponding
+Source.
+
+  The Corresponding Source for a work in source code form is that
+same work.
+
+  2. Basic Permissions.
+
+  All rights granted under this License are granted for the term of
+copyright on the Program, and are irrevocable provided the stated
+conditions are met.  This License explicitly affirms your unlimited
+permission to run the unmodified Program.  The output from running a
+covered work is covered by this License only if the output, given its
+content, constitutes a covered work.  This License acknowledges your
+rights of fair use or other equivalent, as provided by copyright law.
+
+  You may make, run and propagate covered works that you do not
+convey, without conditions so long as your license otherwise remains
+in force.  You may convey covered works to others for the sole purpose
+of having them make modifications exclusively for you, or provide you
+with facilities for running those works, provided that you comply with
+the terms of this License in conveying all material for which you do
+not control copyright.  Those thus making or running the covered works
+for you must do so exclusively on your behalf, under your direction
+and control, on terms that prohibit them from making any copies of
+your copyrighted material outside their relationship with you.
+
+  Conveying under any other circumstances is permitted solely under
+the conditions stated below.  Sublicensing is not allowed; section 10
+makes it unnecessary.
+
+  3. Protecting Users' Legal Rights From Anti-Circumvention Law.
+
+  No covered work shall be deemed part of an effective technological
+measure under any applicable law fulfilling obligations under article
+11 of the WIPO copyright treaty adopted on 20 December 1996, or
+similar laws prohibiting or restricting circumvention of such
+measures.
+
+  When you convey a covered work, you waive any legal power to forbid
+circumvention of technological measures to the extent such circumvention
+is effected by exercising rights under this License with respect to
+the covered work, and you disclaim any intention to limit operation or
+modification of the work as a means of enforcing, against the work's
+users, your or third parties' legal rights to forbid circumvention of
+technological measures.
+
+  4. Conveying Verbatim Copies.
+
+  You may convey verbatim copies of the Program's source code as you
+receive it, in any medium, provided that you conspicuously and
+appropriately publish on each copy an appropriate copyright notice;
+keep intact all notices stating that this License and any
+non-permissive terms added in accord with section 7 apply to the code;
+keep intact all notices of the absence of any warranty; and give all
+recipients a copy of this License along with the Program.
+
+  You may charge any price or no price for each copy that you convey,
+and you may offer support or warranty protection for a fee.
+
+  5. Conveying Modified Source Versions.
+
+  You may convey a work based on the Program, or the modifications to
+produce it from the Program, in the form of source code under the
+terms of section 4, provided that you also meet all of these conditions:
+
+    a) The work must carry prominent notices stating that you modified
+    it, and giving a relevant date.
+
+    b) The work must carry prominent notices stating that it is
+    released under this License and any conditions added under section
+    7.  This requirement modifies the requirement in section 4 to
+    "keep intact all notices".
+
+    c) You must license the entire work, as a whole, under this
+    License to anyone who comes into possession of a copy.  This
+    License will therefore apply, along with any applicable section 7
+    additional terms, to the whole of the work, and all its parts,
+    regardless of how they are packaged.  This License gives no
+    permission to license the work in any other way, but it does not
+    invalidate such permission if you have separately received it.
+
+    d) If the work has interactive user interfaces, each must display
+    Appropriate Legal Notices; however, if the Program has interactive
+    interfaces that do not display Appropriate Legal Notices, your
+    work need not make them do so.
+
+  A compilation of a covered work with other separate and independent
+works, which are not by their nature extensions of the covered work,
+and which are not combined with it such as to form a larger program,
+in or on a volume of a storage or distribution medium, is called an
+"aggregate" if the compilation and its resulting copyright are not
+used to limit the access or legal rights of the compilation's users
+beyond what the individual works permit.  Inclusion of a covered work
+in an aggregate does not cause this License to apply to the other
+parts of the aggregate.
+
+  6. Conveying Non-Source Forms.
+
+  You may convey a covered work in object code form under the terms
+of sections 4 and 5, provided that you also convey the
+machine-readable Corresponding Source under the terms of this License,
+in one of these ways:
+
+    a) Convey the object code in, or embodied in, a physical product
+    (including a physical distribution medium), accompanied by the
+    Corresponding Source fixed on a durable physical medium
+    customarily used for software interchange.
+
+    b) Convey the object code in, or embodied in, a physical product
+    (including a physical distribution medium), accompanied by a
+    written offer, valid for at least three years and valid for as
+    long as you offer spare parts or customer support for that product
+    model, to give anyone who possesses the object code either (1) a
+    copy of the Corresponding Source for all the software in the
+    product that is covered by this License, on a durable physical
+    medium customarily used for software interchange, for a price no
+    more than your reasonable cost of physically performing this
+    conveying of source, or (2) access to copy the
+    Corresponding Source from a network server at no charge.
+
+    c) Convey individual copies of the object code with a copy of the
+    written offer to provide the Corresponding Source.  This
+    alternative is allowed only occasionally and noncommercially, and
+    only if you received the object code with such an offer, in accord
+    with subsection 6b.
+
+    d) Convey the object code by offering access from a designated
+    place (gratis or for a charge), and offer equivalent access to the
+    Corresponding Source in the same way through the same place at no
+    further charge.  You need not require recipients to copy the
+    Corresponding Source along with the object code.  If the place to
+    copy the object code is a network server, the Corresponding Source
+    may be on a different server (operated by you or a third party)
+    that supports equivalent copying facilities, provided you maintain
+    clear directions next to the object code saying where to find the
+    Corresponding Source.  Regardless of what server hosts the
+    Corresponding Source, you remain obligated to ensure that it is
+    available for as long as needed to satisfy these requirements.
+
+    e) Convey the object code using peer-to-peer transmission, provided
+    you inform other peers where the object code and Corresponding
+    Source of the work are being offered to the general public at no
+    charge under subsection 6d.
+
+  A separable portion of the object code, whose source code is excluded
+from the Corresponding Source as a System Library, need not be
+included in conveying the object code work.
+
+  A "User Product" is either (1) a "consumer product", which means any
+tangible personal property which is normally used for personal, family,
+or household purposes, or (2) anything designed or sold for incorporation
+into a dwelling.  In determining whether a product is a consumer product,
+doubtful cases shall be resolved in favor of coverage.  For a particular
+product received by a particular user, "normally used" refers to a
+typical or common use of that class of product, regardless of the status
+of the particular user or of the way in which the particular user
+actually uses, or expects or is expected to use, the product.  A product
+is a consumer product regardless of whether the product has substantial
+commercial, industrial or non-consumer uses, unless such uses represent
+the only significant mode of use of the product.
+
+  "Installation Information" for a User Product means any methods,
+procedures, authorization keys, or other information required to install
+and execute modified versions of a covered work in that User Product from
+a modified version of its Corresponding Source.  The information must
+suffice to ensure that the continued functioning of the modified object
+code is in no case prevented or interfered with solely because
+modification has been made.
+
+  If you convey an object code work under this section in, or with, or
+specifically for use in, a User Product, and the conveying occurs as
+part of a transaction in which the right of possession and use of the
+User Product is transferred to the recipient in perpetuity or for a
+fixed term (regardless of how the transaction is characterized), the
+Corresponding Source conveyed under this section must be accompanied
+by the Installation Information.  But this requirement does not apply
+if neither you nor any third party retains the ability to install
+modified object code on the User Product (for example, the work has
+been installed in ROM).
+
+  The requirement to provide Installation Information does not include a
+requirement to continue to provide support service, warranty, or updates
+for a work that has been modified or installed by the recipient, or for
+the User Product in which it has been modified or installed.  Access to a
+network may be denied when the modification itself materially and
+adversely affects the operation of the network or violates the rules and
+protocols for communication across the network.
+
+  Corresponding Source conveyed, and Installation Information provided,
+in accord with this section must be in a format that is publicly
+documented (and with an implementation available to the public in
+source code form), and must require no special password or key for
+unpacking, reading or copying.
+
+  7. Additional Terms.
+
+  "Additional permissions" are terms that supplement the terms of this
+License by making exceptions from one or more of its conditions.
+Additional permissions that are applicable to the entire Program shall
+be treated as though they were included in this License, to the extent
+that they are valid under applicable law.  If additional permissions
+apply only to part of the Program, that part may be used separately
+under those permissions, but the entire Program remains governed by
+this License without regard to the additional permissions.
+
+  When you convey a copy of a covered work, you may at your option
+remove any additional permissions from that copy, or from any part of
+it.  (Additional permissions may be written to require their own
+removal in certain cases when you modify the work.)  You may place
+additional permissions on material, added by you to a covered work,
+for which you have or can give appropriate copyright permission.
+
+  Notwithstanding any other provision of this License, for material you
+add to a covered work, you may (if authorized by the copyright holders of
+that material) supplement the terms of this License with terms:
+
+    a) Disclaiming warranty or limiting liability differently from the
+    terms of sections 15 and 16 of this License; or
+
+    b) Requiring preservation of specified reasonable legal notices or
+    author attributions in that material or in the Appropriate Legal
+    Notices displayed by works containing it; or
+
+    c) Prohibiting misrepresentation of the origin of that material, or
+    requiring that modified versions of such material be marked in
+    reasonable ways as different from the original version; or
+
+    d) Limiting the use for publicity purposes of names of licensors or
+    authors of the material; or
+
+    e) Declining to grant rights under trademark law for use of some
+    trade names, trademarks, or service marks; or
+
+    f) Requiring indemnification of licensors and authors of that
+    material by anyone who conveys the material (or modified versions of
+    it) with contractual assumptions of liability to the recipient, for
+    any liability that these contractual assumptions directly impose on
+    those licensors and authors.
+
+  All other non-permissive additional terms are considered "further
+restrictions" within the meaning of section 10.  If the Program as you
+received it, or any part of it, contains a notice stating that it is
+governed by this License along with a term that is a further
+restriction, you may remove that term.  If a license document contains
+a further restriction but permits relicensing or conveying under this
+License, you may add to a covered work material governed by the terms
+of that license document, provided that the further restriction does
+not survive such relicensing or conveying.
+
+  If you add terms to a covered work in accord with this section, you
+must place, in the relevant source files, a statement of the
+additional terms that apply to those files, or a notice indicating
+where to find the applicable terms.
+
+  Additional terms, permissive or non-permissive, may be stated in the
+form of a separately written license, or stated as exceptions;
+the above requirements apply either way.
+
+  8. Termination.
+
+  You may not propagate or modify a covered work except as expressly
+provided under this License.  Any attempt otherwise to propagate or
+modify it is void, and will automatically terminate your rights under
+this License (including any patent licenses granted under the third
+paragraph of section 11).
+
+  However, if you cease all violation of this License, then your
+license from a particular copyright holder is reinstated (a)
+provisionally, unless and until the copyright holder explicitly and
+finally terminates your license, and (b) permanently, if the copyright
+holder fails to notify you of the violation by some reasonable means
+prior to 60 days after the cessation.
+
+  Moreover, your license from a particular copyright holder is
+reinstated permanently if the copyright holder notifies you of the
+violation by some reasonable means, this is the first time you have
+received notice of violation of this License (for any work) from that
+copyright holder, and you cure the violation prior to 30 days after
+your receipt of the notice.
+
+  Termination of your rights under this section does not terminate the
+licenses of parties who have received copies or rights from you under
+this License.  If your rights have been terminated and not permanently
+reinstated, you do not qualify to receive new licenses for the same
+material under section 10.
+
+  9. Acceptance Not Required for Having Copies.
+
+  You are not required to accept this License in order to receive or
+run a copy of the Program.  Ancillary propagation of a covered work
+occurring solely as a consequence of using peer-to-peer transmission
+to receive a copy likewise does not require acceptance.  However,
+nothing other than this License grants you permission to propagate or
+modify any covered work.  These actions infringe copyright if you do
+not accept this License.  Therefore, by modifying or propagating a
+covered work, you indicate your acceptance of this License to do so.
+
+  10. Automatic Licensing of Downstream Recipients.
+
+  Each time you convey a covered work, the recipient automatically
+receives a license from the original licensors, to run, modify and
+propagate that work, subject to this License.  You are not responsible
+for enforcing compliance by third parties with this License.
+
+  An "entity transaction" is a transaction transferring control of an
+organization, or substantially all assets of one, or subdividing an
+organization, or merging organizations.  If propagation of a covered
+work results from an entity transaction, each party to that
+transaction who receives a copy of the work also receives whatever
+licenses to the work the party's predecessor in interest had or could
+give under the previous paragraph, plus a right to possession of the
+Corresponding Source of the work from the predecessor in interest, if
+the predecessor has it or can get it with reasonable efforts.
+
+  You may not impose any further restrictions on the exercise of the
+rights granted or affirmed under this License.  For example, you may
+not impose a license fee, royalty, or other charge for exercise of
+rights granted under this License, and you may not initiate litigation
+(including a cross-claim or counterclaim in a lawsuit) alleging that
+any patent claim is infringed by making, using, selling, offering for
+sale, or importing the Program or any portion of it.
+
+  11. Patents.
+
+  A "contributor" is a copyright holder who authorizes use under this
+License of the Program or a work on which the Program is based.  The
+work thus licensed is called the contributor's "contributor version".
+
+  A contributor's "essential patent claims" are all patent claims
+owned or controlled by the contributor, whether already acquired or
+hereafter acquired, that would be infringed by some manner, permitted
+by this License, of making, using, or selling its contributor version,
+but do not include claims that would be infringed only as a
+consequence of further modification of the contributor version.  For
+purposes of this definition, "control" includes the right to grant
+patent sublicenses in a manner consistent with the requirements of
+this License.
+
+  Each contributor grants you a non-exclusive, worldwide, royalty-free
+patent license under the contributor's essential patent claims, to
+make, use, sell, offer for sale, import and otherwise run, modify and
+propagate the contents of its contributor version.
+
+  In the following three paragraphs, a "patent license" is any express
+agreement or commitment, however denominated, not to enforce a patent
+(such as an express permission to practice a patent or covenant not to
+sue for patent infringement).  To "grant" such a patent license to a
+party means to make such an agreement or commitment not to enforce a
+patent against the party.
+
+  If you convey a covered work, knowingly relying on a patent license,
+and the Corresponding Source of the work is not available for anyone
+to copy, free of charge and under the terms of this License, through a
+publicly available network server or other readily accessible means,
+then you must either (1) cause the Corresponding Source to be so
+available, or (2) arrange to deprive yourself of the benefit of the
+patent license for this particular work, or (3) arrange, in a manner
+consistent with the requirements of this License, to extend the patent
+license to downstream recipients.  "Knowingly relying" means you have
+actual knowledge that, but for the patent license, your conveying the
+covered work in a country, or your recipient's use of the covered work
+in a country, would infringe one or more identifiable patents in that
+country that you have reason to believe are valid.
+
+  If, pursuant to or in connection with a single transaction or
+arrangement, you convey, or propagate by procuring conveyance of, a
+covered work, and grant a patent license to some of the parties
+receiving the covered work authorizing them to use, propagate, modify
+or convey a specific copy of the covered work, then the patent license
+you grant is automatically extended to all recipients of the covered
+work and works based on it.
+
+  A patent license is "discriminatory" if it does not include within
+the scope of its coverage, prohibits the exercise of, or is
+conditioned on the non-exercise of one or more of the rights that are
+specifically granted under this License.  You may not convey a covered
+work if you are a party to an arrangement with a third party that is
+in the business of distributing software, under which you make payment
+to the third party based on the extent of your activity of conveying
+the work, and under which the third party grants, to any of the
+parties who would receive the covered work from you, a discriminatory
+patent license (a) in connection with copies of the covered work
+conveyed by you (or copies made from those copies), or (b) primarily
+for and in connection with specific products or compilations that
+contain the covered work, unless you entered into that arrangement,
+or that patent license was granted, prior to 28 March 2007.
+
+  Nothing in this License shall be construed as excluding or limiting
+any implied license or other defenses to infringement that may
+otherwise be available to you under applicable patent law.
+
+  12. No Surrender of Others' Freedom.
+
+  If conditions are imposed on you (whether by court order, agreement or
+otherwise) that contradict the conditions of this License, they do not
+excuse you from the conditions of this License.  If you cannot convey a
+covered work so as to satisfy simultaneously your obligations under this
+License and any other pertinent obligations, then as a consequence you may
+not convey it at all.  For example, if you agree to terms that obligate you
+to collect a royalty for further conveying from those to whom you convey
+the Program, the only way you could satisfy both those terms and this
+License would be to refrain entirely from conveying the Program.
+
+  13. Remote Network Interaction; Use with the GNU General Public License.
+
+  Notwithstanding any other provision of this License, if you modify the
+Program, your modified version must prominently offer all users
+interacting with it remotely through a computer network (if your version
+supports such interaction) an opportunity to receive the Corresponding
+Source of your version by providing access to the Corresponding Source
+from a network server at no charge, through some standard or customary
+means of facilitating copying of software.  This Corresponding Source
+shall include the Corresponding Source for any work covered by version 3
+of the GNU General Public License that is incorporated pursuant to the
+following paragraph.
+
+  Notwithstanding any other provision of this License, you have
+permission to link or combine any covered work with a work licensed
+under version 3 of the GNU General Public License into a single
+combined work, and to convey the resulting work.  The terms of this
+License will continue to apply to the part which is the covered work,
+but the work with which it is combined will remain governed by version
+3 of the GNU General Public License.
+
+  14. Revised Versions of this License.
+
+  The Free Software Foundation may publish revised and/or new versions of
+the GNU Affero General Public License from time to time.  Such new versions
+will be similar in spirit to the present version, but may differ in detail to
+address new problems or concerns.
+
+  Each version is given a distinguishing version number.  If the
+Program specifies that a certain numbered version of the GNU Affero General
+Public License "or any later version" applies to it, you have the
+option of following the terms and conditions either of that numbered
+version or of any later version published by the Free Software
+Foundation.  If the Program does not specify a version number of the
+GNU Affero General Public License, you may choose any version ever published
+by the Free Software Foundation.
+
+  If the Program specifies that a proxy can decide which future
+versions of the GNU Affero General Public License can be used, that proxy's
+public statement of acceptance of a version permanently authorizes you
+to choose that version for the Program.
+
+  Later license versions may give you additional or different
+permissions.  However, no additional obligations are imposed on any
+author or copyright holder as a result of your choosing to follow a
+later version.
+
+  15. Disclaimer of Warranty.
+
+  THERE IS NO WARRANTY FOR THE PROGRAM, TO THE EXTENT PERMITTED BY
+APPLICABLE LAW.  EXCEPT WHEN OTHERWISE STATED IN WRITING THE COPYRIGHT
+HOLDERS AND/OR OTHER PARTIES PROVIDE THE PROGRAM "AS IS" WITHOUT WARRANTY
+OF ANY KIND, EITHER EXPRESSED OR IMPLIED, INCLUDING, BUT NOT LIMITED TO,
+THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
+PURPOSE.  THE ENTIRE RISK AS TO THE QUALITY AND PERFORMANCE OF THE PROGRAM
+IS WITH YOU.  SHOULD THE PROGRAM PROVE DEFECTIVE, YOU ASSUME THE COST OF
+ALL NECESSARY SERVICING, REPAIR OR CORRECTION.
+
+  16. Limitation of Liability.
+
+  IN NO EVENT UNLESS REQUIRED BY APPLICABLE LAW OR AGREED TO IN WRITING
+WILL ANY COPYRIGHT HOLDER, OR ANY OTHER PARTY WHO MODIFIES AND/OR CONVEYS
+THE PROGRAM AS PERMITTED ABOVE, BE LIABLE TO YOU FOR DAMAGES, INCLUDING ANY
+GENERAL, SPECIAL, INCIDENTAL OR CONSEQUENTIAL DAMAGES ARISING OUT OF THE
+USE OR INABILITY TO USE THE PROGRAM (INCLUDING BUT NOT LIMITED TO LOSS OF
+DATA OR DATA BEING RENDERED INACCURATE OR LOSSES SUSTAINED BY YOU OR THIRD
+PARTIES OR A FAILURE OF THE PROGRAM TO OPERATE WITH ANY OTHER PROGRAMS),
+EVEN IF SUCH HOLDER OR OTHER PARTY HAS BEEN ADVISED OF THE POSSIBILITY OF
+SUCH DAMAGES.
+
+  17. Interpretation of Sections 15 and 16.
+
+  If the disclaimer of warranty and limitation of liability provided
+above cannot be given local legal effect according to their terms,
+reviewing courts shall apply local law that most closely approximates
+an absolute waiver of all civil liability in connection with the
+Program, unless a warranty or assumption of liability accompanies a
+copy of the Program in return for a fee.
+
+                     END OF TERMS AND CONDITIONS
+
+            How to Apply These Terms to Your New Programs
+
+  If you develop a new program, and you want it to be of the greatest
+possible use to the public, the best way to achieve this is to make it
+free software which everyone can redistribute and change under these terms.
+
+  To do so, attach the following notices to the program.  It is safest
+to attach them to the start of each source file to most effectively
+state the exclusion of warranty; and each file should have at least
+the "copyright" line and a pointer to where the full notice is found.
+
+    <one line to give the program's name and a brief idea of what it does.>
+    Copyright (C) <year>  <name of author>
+
+    This program is free software: you can redistribute it and/or modify
+    it under the terms of the GNU Affero General Public License as published
+    by the Free Software Foundation, either version 3 of the License, or
+    (at your option) any later version.
+
+    This program is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+    GNU Affero General Public License for more details.
+
+    You should have received a copy of the GNU Affero General Public License
+    along with this program.  If not, see <https://www.gnu.org/licenses/>.
+
+Also add information on how to contact you by electronic and paper mail.
+
+  If your software can interact with users remotely through a computer
+network, you should also make sure that it provides a way for users to
+get its source.  For example, if your program is a web application, its
+interface could display a "Source" link that leads users to an archive
+of the code.  There are many ways you could offer source, and different
+solutions will be better for different programs; see section 13 for the
+specific requirements.
+
+  You should also get your employer (if you work as a programmer) or school,
+if any, to sign a "copyright disclaimer" for the program, if necessary.
+For more information on this, and how to apply and follow the GNU AGPL, see
+<https://www.gnu.org/licenses/>.

BanG-Dream-VITS/README.md

@@ -0,0 +1,15 @@
+---
+title: BanG Dream! It's MyGO!!!!!
+emoji: 🎸
+colorFrom: green
+colorTo: gray
+sdk: gradio
+sdk_version: 4.12.0
+app_file: app.py
+pinned: true
+python_version: '3.10'
+short_description: Generate Voice BanG Dream! VITS
+---
+
+# BanGDream MyGO VITS
+Aplikasi ini dijalankan dengan Python 3.10 untuk stabilitas library audio.

BanG-Dream-VITS/app.py

@@ -0,0 +1,329 @@
+import os
+os.system("pip uninstall -y gradio gradio-client")
+os.system("pip install gradio==3.50.2 huggingface-hub==0.25.2 transformers==4.44.2")
+import gradio as gr
+import logging
+import torch
+import numpy as np
+import random
+import shutil
+import gc
+import psutil
+import re
+from huggingface_hub import snapshot_download
+
+# --- 0. AUTO CLEANER FUNCTION ---
+def cleanup_memory():
+    gc.collect()
+    if torch.cuda.is_available():
+        torch.cuda.empty_cache()
+    temp_dir = "temp"
+    if os.path.exists(temp_dir):
+        try:
+            shutil.rmtree(temp_dir)
+            os.makedirs(temp_dir)
+        except: pass
+
+cleanup_memory()
+
+# --- 1. SETUP ASSETS ---
+def setup_shittim_box():
+    repo_id = "Plana-Archive/Anime-Models"
+    # Menyesuaikan dengan path subfolder yang benar di Hugging Face
+    remote_sub_folder = "Anime-VITS/MyGO-AveMujica-VITS"
+    
+    if not os.path.exists("Data"):
+        try:
+            # Mendownload konten secara spesifik berdasarkan pola subfolder repositori
+            snapshot_download(
+                repo_id=repo_id, 
+                allow_patterns=[f"{remote_sub_folder}/**/*"], 
+                local_dir="."
+            )
+            
+            # Memindahkan folder utama yang dibutuhkan aplikasi ke root direktori
+            folders = ["Data", "bert", "emotional", "logs", "image"]
+            for f in folders:
+                src = os.path.join(remote_sub_folder, f)
+                if os.path.exists(src):
+                    if os.path.exists(f): 
+                        shutil.rmtree(f)
+                    shutil.move(src, f)
+            
+            # Membersihkan sisa folder Anime-VITS hasil download setelah dipindahkan
+            if os.path.exists("Anime-VITS"): 
+                shutil.rmtree("Anime-VITS")
+        except Exception: 
+            pass
+
+setup_shittim_box()
+
+# --- 2. IMPORT MODUL ---
+import utils
+import commons
+from models import SynthesizerTrn
+from text.symbols import symbols
+from text import cleaned_text_to_sequence, get_bert
+from text.cleaner import clean_text
+
+logging.basicConfig(level=logging.INFO)
+device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+
+# --- 3. CONFIG & CHARACTER LIST ---
+config_path = "Data/BangDream/config.json"
+hps = utils.get_hparams_from_file(config_path)
+model_dir = "Data/BangDream/models/"
+model_file = [os.path.join(model_dir, f) for f in os.listdir(model_dir) if f.endswith('.pth')][-1]
+
+net_g = SynthesizerTrn(
+    len(symbols), hps.data.filter_length // 2 + 1,
+    hps.train.segment_size // hps.data.hop_length,
+    n_speakers=hps.data.n_speakers, **hps.model
+).to(device)
+_ = net_g.eval()
+_ = utils.load_checkpoint(model_file, net_g, None, skip_optimizer=True)
+
+NAME_MAP = {
+    "燈": "Tomori", "そよ": "Soyo",
+    "睦": "Mutsumi", "祥子": "Sakiko",
+    "愛音": "Anon", "立希": "Taki",
+    "楽奈": "Rana", "海鈴": "Umiri",
+    "初華": "Uika", "にゃむ": "Nyaamu"
+}
+
+additional_characters = [
+    "あこ", "こころ", "たえ", "つくし", "つぐみ", "はぐみ", "ひまり", "ましろ", "ますき", "りみ",
+    "イヴ", "チュチュ", "パレオ", "モカ", "リサ", "レイヤ", "ロック", "一愛", "七深", "光",
+    "克洛迪娜", "八千代", "千聖", "友希那", "司", "壘", "奈奈", "巴", "彩", "悠悠子", "文",
+    "日菜", "晶", "有咲", "未知留", "栞", "沙綾", "珠緒", "瑠唯", "真晝", "真矢", "純那",
+    "美咲", "美帆", "美空", "艾露", "艾露露", "花音", "菈樂菲", "華戀", "薫", "蘭", "透子",
+    "雙葉", "靜羽", "香子", "香澄", "麻弥"
+]
+
+additional_romaji = {
+    "あこ": "Ako", "こころ": "Kokoro", "たえ": "Tae", "つくし": "Tsukushi", "つぐみ": "Tsugumi",
+    "はぐみ": "Hagumi", "ひまり": "Himari", "ましろ": "Mashiro", "ますき": "Masuki", "りみ": "Rimi",
+    "イヴ": "Eve", "チュチュ": "Chuchu", "パレオ": "Pareo", "モカ": "Moka", "リサ": "Lisa",
+    "レイヤ": "Layer", "ロック": "Rock", "一愛": "Kazusa", "七深": "Nanami", "光": "Hikaru",
+    "克洛迪娜": "Claudine", "八千代": "Yachiyo", "千聖": "Chisato", "友希那": "Yukina", "司": "Tsukasa",
+    "壘": "Rui", "奈奈": "Nana", "巴": "Tomoe", "彩": "Aya", "悠悠子": "Yuyuko", "文": "Fumi",
+    "日菜": "Hina", "晶": "Akira", "有咲": "Arisa", "未知留": "Michiru", "栞": "Shiori",
+    "沙綾": "Saaya", "珠緒": "Tamao", "瑠唯": "Rui", "真晝": "Mahiru", "真矢": "Maya",
+    "純那": "Junna", "美咲": "Misaki", "美帆": "Miho", "美空": "Misora", "艾露": "Airu",
+    "艾露露": "Airuru", "花音": "Kanon", "菈樂菲": "Rarafe", "華戀": "Karen", "薫": "Kaoru",
+    "蘭": "Ran", "透子": "Touko", "雙葉": "Futaba", "靜羽": "Shizuha", "香子": "Kyouko",
+    "香澄": "Kasumi", "麻弥": "Maya"
+}
+
+for char, rom in additional_romaji.items():
+    if char not in NAME_MAP:
+        NAME_MAP[char] = rom
+
+original_spk = ["燈", "そよ", "睦", "祥子", "愛音", "立希", "���奈", "海鈴", "初華", "にゃむ"]
+new_spk = []
+for char in additional_characters:
+    if char not in original_spk:
+        new_spk.append(char)
+SORTED_SPK = original_spk + new_spk
+
+# --- 4. INFERENCE ENGINE ---
+def infer(text, sdp_ratio, noise_scale, noise_scale_w, length_scale, sid_name):
+    if not text.strip(): return "Teks kosong!", None
+    try:
+        if psutil.virtual_memory().percent > 85:
+            cleanup_memory()
+
+        norm_text, phone, tone, word2ph = clean_text(text, "JP")
+        phone, tone, lang_seq = cleaned_text_to_sequence(phone, tone, "JP")
+        if hps.data.add_blank:
+            phone = commons.intersperse(phone, 0)
+            tone = commons.intersperse(tone, 0)
+            lang_seq = commons.intersperse(lang_seq, 0)
+            for i in range(len(word2ph)): word2ph[i] = word2ph[i] * 2
+            word2ph[0] += 1
+        bert_ori = get_bert(norm_text, word2ph, "JP", device, "", 0.7)
+        with torch.no_grad():
+            x = torch.LongTensor(phone).to(device).unsqueeze(0)
+            t = torch.LongTensor(tone).to(device).unsqueeze(0)
+            l = torch.LongTensor(lang_seq).to(device).unsqueeze(0)
+            bert = torch.zeros(1024, len(phone)).to(device).unsqueeze(0).float()
+            ja_bert = bert_ori.to(device).unsqueeze(0).float()
+            en_bert = torch.zeros(1024, len(phone)).to(device).unsqueeze(0).float()
+            sid_id = torch.LongTensor([hps.data.spk2id[sid_name]]).to(device)
+            audio = net_g.infer(x, torch.LongTensor([len(phone)]).to(device), sid_id, t, l, bert, ja_bert, en_bert,
+                                sdp_ratio=sdp_ratio, noise_scale=noise_scale, noise_scale_w=noise_scale_w,
+                                length_scale=length_scale)[0][0,0].data.cpu().float().numpy()
+        cleanup_memory()
+        return "Berhasil Membuat ✅!", (hps.data.sampling_rate, (audio * 32767).astype(np.int16))
+    except Exception as e:
+        cleanup_memory()
+        return f"Error: {str(e)}", None
+
+def get_char_info_html():
+    html = f'''<div class="info-content-area" id="info-content-main" style="display: grid; grid-template-columns: 1fr 1fr; gap: 10px; padding: 15px; background: white; border: 1px solid #e1f0e5; border-top: none; border-radius: 0 0 10px 10px; max-height: 300px; overflow-y: auto;">'''
+    for name in SORTED_SPK:
+        romaji = NAME_MAP.get(name, "Character")
+        html += f'''
+        <div style="border: 1px solid #e1f0e5; border-radius: 12px; padding: 10px; border-left: 5px solid #a8e6cf; background: #fff;">
+            <div style="font-weight: 800; color: #2d3748; font-size: 14px; margin-bottom: 2px;">{name}</div>
+            <div style="color: #a0aec0; font-size: 11px;">{romaji}</div>
+        </div>
+        '''
+    html += "</div>"
+    return html
+
+# --- 5. UI STYLE (BABY GREEN THEME) ---
+css = """
+:root { 
+    --primary-600: #a8e6cf !important; 
+    --accent-600: #a8e6cf !important;
+}
+.gradio-container, .gradio-container * {
+    --loader-color: #dcedc1 !important;
+}
+.ba-header-container {
+    border: 1.5px solid #e1f0e5;
+    border-radius: 12px;
+    margin-bottom: 12px;
+    background: white;
+    overflow: hidden;
+    line-height: 0;
+}
+.ba-header-container img {
+    width: 100%;
+    height: auto;
+}
+.status-container {
+    border: 1.5px solid #e1f0e5;
+    border-radius: 12px;
+    padding: 15px 22px;
+    margin-bottom: 20px;
+    background: white;
+}
+.status-title {
+    color: #88d498 !important;
+    font-weight: 800;
+    font-size: 16px;
+    margin-bottom: 8px;
+}
+.dot { 
+    height: 8px; 
+    width: 8px; 
+    background-color: #88d498; 
+    border-radius: 50%; 
+    display: inline-block; 
+    margin-right: 6px; 
+    box-shadow: 0 0 0 0 rgba(136, 212, 152, 1);
+    animation: pulse-green 2s infinite; 
+}
+@keyframes pulse-green {
+  0% { transform: scale(0.95); box-shadow: 0 0 0 0 rgba(136, 212, 152, 0.7); }
+  70% { transform: scale(1); box-shadow: 0 0 0 10px rgba(136, 212, 152, 0); }
+  100% { transform: scale(0.95); box-shadow: 0 0 0 0 rgba(136, 212, 152, 0); }
+}
+
+.text-green-bold { color: #88d498 !important; font-weight: 900 !important; }
+.text-blue-status { color: #88d498 !important; }
+.slim-card { max-width: 480px; margin: 0 auto; background: transparent; padding: 10px; }
+.char-img { border-radius: 15px; border: 1px solid #e1f0e5; margin-bottom: 15px; width: 100%; }
+.scroll-box { height: 220px; overflow-y: auto; border: 1px solid #f0f4f8; border-radius: 12px; padding: 10px; background: #fafbfc; margin-bottom: 10px; }
+.char-btn { background: white !important; border: 1px solid #e2e8f0 !important; border-left: 5px solid #a8e6cf !important; text-align: left !important; padding: 8px !important; font-size: 12px !important; margin-bottom: 4px !important; width: 100%; color: #4a5568 !important; }
+.warning-card { background: #fffcf0; border: 2px dashed #ffd3b6; border-radius: 10px; padding: 12px; margin-bottom: 15px; text-align: center; }
+
+.jp-btn { 
+    background: #a8e6cf !important; 
+    border: 1px solid #a8e6cf !important; 
+    color: #ffffff !important; 
+    font-weight: 700 !important; 
+    border-radius: 10px !important; 
+    margin-bottom: 10px; 
+    font-size: 12px !important; 
+    width: 100%; 
+    cursor: pointer;
+}
+
+.gen-btn { background: #a8e6cf !important; color: #ffffff !important; font-weight: 700 !important; border-radius: 12px !important; height: 45px !important; width: 100%; border: none !important; cursor: pointer; }
+.info-header-custom { background: #dcedc1 !important; color: #4a5568 !important; border: none !important; border-radius: 8px 8px 0 0 !important; padding: 12px 15px !important; width: 100% !important; cursor: pointer; font-weight: 800 !important; margin-top: 5px; }
+.credit-footer { margin-top: 25px; padding: 15px; background: white; border-radius: 12px; text-align: center; border-bottom: 4px solid #a8e6cf; color: #94a3b8; font-weight: 700; font-size: 12px; letter-spacing: 2px; }
+
+.gif-container { display: flex; justify-content: center; margin-top: 15px; margin-bottom: 10px; width: 100%; }
+.gif-container img { border-radius: 12px; max-width: 100%; height: auto; border: 1.5px solid #e1f0e5; object-fit: contain; }
+
+label[for="input_text"] {
+    background: #a8e6cf !important;
+    color: #4a5568 !important;
+    padding: 6px 12px !important;
+    border-radius: 12px !important;
+    font-weight: 600 !important;
+    display: inline-block !important;
+    width: auto !important;
+    margin-bottom: 6px !important;
+}
+"""
+
+with gr.Blocks(css=css) as app:
+    with gr.Column(elem_classes="slim-card"):
+        gr.HTML("""
+            <div class="ba-header-container">
+                <img src="https://huggingface.co/spaces/Plana-Archive/BanG-Dream-VITS/resolve/main/BANNER.PNG" alt="BanG Dream Banner">
+            </div>
+            <div class="status-container">
+                <div class="status-title">System Status</div>
+                <div class="status-item"><span style="color:#4a5568">Model :</span> <span class="text-green-bold">&nbsp;LOADED ✅</span></div>
+                <div class="status-item"><span style="color:#88d498; font-weight:700;"><span class="dot"></span> Online</span></div>
+            </div>
+        """)
+        
+        char_img = gr.Image(value=f"image/{SORTED_SPK[0]}.png", show_label=False, interactive=False, height=300, elem_classes="char-img")
+        active_spk = gr.State(SORTED_SPK[0])
+        char_display = gr.Markdown(f"📍 Selected: **{SORTED_SPK[0]}**")
+
+        with gr.Column(elem_classes="scroll-box"):
+            for name in SORTED_SPK:
+                romaji = NAME_MAP.get(name, "Character")
+                btn = gr.Button(f"👤 {name} | {romaji}", elem_classes="char-btn")
+                btn.click(lambda n=name: (n, f"📍 Selected: **{n}**", f"image/{n}.png"), 
+                          outputs=[active_spk, char_display, char_img])
+
+        gr.HTML("""
+            <div class="warning-card">
+                <div style="color:#f5a623;font-weight:800;font-size:13px;">🔖 PERINGATAN MINNA 🔖</div>
+                <div style="color:#855d1a;font-size:11px;font-weight:600;">
+                    Pilih Karakter, masukkan teks Jepang, lalu klik Generate Voice! ✨
+                </div>
+            </div>
+        """)
+
+        input_text = gr.TextArea(label="Input Text (Japanese)", value="こんにちは、お会いできて光栄です。", lines=3, elem_id="input_text")
+        
+        def get_random_jp_text():
+            texts = ["こんにちは！", "私の音楽 を聴いてください。", "一生、離さないから。", "私たちは、まだ終わっていない。", "Ave Mujicaの世界へようこそ。"]
+            return random.choice(texts)
+            
+        gr.Button("🍏 INPUTS RANDOM TEXT 🍏", elem_classes="jp-btn").click(get_random_jp_text, outputs=[input_text])
+
+        with gr.Row():
+            speed_sl = gr.Slider(label="Speed", minimum=0.5, maximum=2.0, value=1.0, step=0.1)
+            sdp_sl = gr.Slider(label="SDP", minimum=0, maximum=1.0, value=0.2, step=0.1)
+
+        btn_gen = gr.Button("🥒 GENERATE VOICE 🥒", elem_classes="gen-btn")
+        audio_out = gr.Audio(label="Voice Output", type="numpy")
+        status_msg = gr.Textbox(visible=False)
+
+        gr.HTML(f"""
+            <div class="gif-container">
+                <img src="https://huggingface.co/spaces/Plana-Archive/BanG-Dream-VITS/resolve/main/ave-mujica-avemuji.gif" alt="BanG Dream GIF">
+            </div>
+            <button onclick="document.getElementById('info-content-main').style.display = (document.getElementById('info-content-main').style.display === 'none') ? 'grid' : 'none';" class="info-header-custom">
+                📑 Character Information 📑
+            </button>
+        """)
+        gr.HTML(get_char_info_html())
+
+        gr.HTML("""<div class="credit-footer">🍏 CREATED BY MUTSUMI 🍏</div>""")
+
+    btn_gen.click(infer, inputs=[input_text, sdp_sl, gr.State(0.6), gr.State(0.8), speed_sl, active_spk],
+                  outputs=[status_msg, audio_out])
+
+if __name__ == "__main__":
+    app.queue().launch(server_name="0.0.0.0", server_port=7860, quiet=True)

BanG-Dream-VITS/app.py (Cadangan utama)

@@ -0,0 +1,260 @@
+import os
+os.system("pip uninstall -y gradio gradio-client")
+os.system("pip install gradio==3.50.2 huggingface-hub==0.25.2 transformers==4.44.2")
+import gradio as gr
+import logging
+import torch
+import numpy as np
+import random
+import shutil
+import gc
+import psutil
+from huggingface_hub import snapshot_download
+
+# --- 0. AUTO CLEANER FUNCTION ---
+def cleanup_memory():
+    gc.collect()
+    if torch.cuda.is_available():
+        torch.cuda.empty_cache()
+    temp_dir = "temp"
+    if os.path.exists(temp_dir):
+        try:
+            shutil.rmtree(temp_dir)
+            os.makedirs(temp_dir)
+        except: pass
+
+cleanup_memory()
+
+# --- 1. SETUP ASSETS ---
+def setup_shittim_box():
+    repo_id = "Plana-Archive/Plana-TTS"
+    sub_folder = "MyGO-AveMujica-VITS"
+    if not os.path.exists("Data"):
+        try:
+            snapshot_download(repo_id=repo_id, allow_patterns=[f"{sub_folder}/**/*"], local_dir=".")
+            folders = ["Data", "bert", "emotional", "logs", "image"]
+            for f in folders:
+                src = os.path.join(sub_folder, f)
+                if os.path.exists(src):
+                    if os.path.exists(f): shutil.rmtree(f)
+                    shutil.move(src, f)
+            if os.path.exists(sub_folder): shutil.rmtree(sub_folder)
+        except Exception: pass
+
+setup_shittim_box()
+
+# --- 2. IMPORT MODUL ---
+import utils
+import commons
+from models import SynthesizerTrn
+from text.symbols import symbols
+from text import cleaned_text_to_sequence, get_bert
+from text.cleaner import clean_text
+
+logging.basicConfig(level=logging.INFO)
+device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+
+# --- 3. CONFIG & CHARACTER LIST ---
+config_path = "Data/BangDream/config.json"
+hps = utils.get_hparams_from_file(config_path)
+model_dir = "Data/BangDream/models/"
+model_file = [os.path.join(model_dir, f) for f in os.listdir(model_dir) if f.endswith('.pth')][-1]
+
+net_g = SynthesizerTrn(
+    len(symbols), hps.data.filter_length // 2 + 1, 
+    hps.train.segment_size // hps.data.hop_length, 
+    n_speakers=hps.data.n_speakers, **hps.model
+).to(device)
+_ = net_g.eval()
+_ = utils.load_checkpoint(model_file, net_g, None, skip_optimizer=True)
+
+NAME_MAP = {
+    "燈": "Tomori", "そよ": "Soyo", 
+    "睦": "Mutsumi", "祥子": "Sakiko",
+    "愛音": "Anon", "立希": "Taki",
+    "楽奈": "Rana", "海鈴": "Umiri", 
+    "初華": "Uika", "にゃむ": "Nyaamu"
+}
+SORTED_SPK = ["燈", "そよ", "睦", "祥子", "愛音", "立希", "楽奈", "海鈴", "初華", "にゃむ"]
+
+# --- 4. INFERENCE ENGINE ---
+def infer(text, sdp_ratio, noise_scale, noise_scale_w, length_scale, sid_name):
+    if not text.strip(): return "Teks kosong!", None
+    try:
+        if psutil.virtual_memory().percent > 85:
+            cleanup_memory()
+
+        norm_text, phone, tone, word2ph = clean_text(text, "JP")
+        phone, tone, lang_seq = cleaned_text_to_sequence(phone, tone, "JP")
+        if hps.data.add_blank:
+            phone = commons.intersperse(phone, 0)
+            tone = commons.intersperse(tone, 0)
+            lang_seq = commons.intersperse(lang_seq, 0)
+            for i in range(len(word2ph)): word2ph[i] = word2ph[i] * 2
+            word2ph[0] += 1
+        bert_ori = get_bert(norm_text, word2ph, "JP", device, "", 0.7)
+        with torch.no_grad():
+            x = torch.LongTensor(phone).to(device).unsqueeze(0)
+            t = torch.LongTensor(tone).to(device).unsqueeze(0)
+            l = torch.LongTensor(lang_seq).to(device).unsqueeze(0)
+            bert = torch.zeros(1024, len(phone)).to(device).unsqueeze(0).float()
+            ja_bert = bert_ori.to(device).unsqueeze(0).float()
+            en_bert = torch.zeros(1024, len(phone)).to(device).unsqueeze(0).float()
+            sid_id = torch.LongTensor([hps.data.spk2id[sid_name]]).to(device)
+            audio = net_g.infer(x, torch.LongTensor([len(phone)]).to(device), sid_id, t, l, bert, ja_bert, en_bert,
+                                sdp_ratio=sdp_ratio, noise_scale=noise_scale, noise_scale_w=noise_scale_w,
+                                length_scale=length_scale)[0][0,0].data.cpu().float().numpy()
+        cleanup_memory()
+        return "Berhasil Membuat ✅!", (hps.data.sampling_rate, (audio * 32767).astype(np.int16))
+    except Exception as e:
+        cleanup_memory()
+        return f"Error: {str(e)}", None
+
+# --- 5. UI DESIGN (Gradio 3.50.2) ---
+css = """
+@import url('https://fonts.googleapis.com/css2?family=Inter:wght@400;700;800&display=swap');
+body, .gradio-container { background-color: #f7f9fc !important; font-family: 'Inter', sans-serif !important; }
+.slim-card { max-width: 480px; margin: 0 auto !important; background: white; border-radius: 20px; padding: 25px; box-shadow: 0 4px 25px rgba(0,0,0,0.03); }
+
+:root { --loader-color: #1299ff !important; }
+.progress-bar, .primary-500, .accent-500 { background-color: #1299ff !important; }
+div[role="progressbar"] > div { background-color: #1299ff !important; }
+svg[class*="loading"] { fill: #1299ff !important; stroke: #1299ff !important; }
+
+.ba-header { text-align: center; margin-bottom: 20px; }
+.ba-header h1 { color: #1299ff !important; font-weight: 800; margin: 0; font-size: 1.6rem; }
+.ba-header p { color: #8a99af; font-size: 10px; letter-spacing: 2px; margin: 0; }
+
+.status-online { background: #f0f7ff; border: 1px solid #d0e4ff; border-radius: 12px; padding: 12px; margin-bottom: 20px; }
+.dot { height: 8px; width: 8px; background-color: #2ecc71; border-radius: 50%; display: inline-block; margin-right: 6px; animation: blink 1.5s infinite; }
+@keyframes blink { 0% { opacity: 1; } 50% { opacity: 0.4; } 100% { opacity: 1; } }
+
+.char-img { border-radius: 15px; border: 1px solid #f1f5f9; margin-bottom: 15px; }
+.warning-minna { background: #fff9f0; border: 1px dashed #ff9800; border-radius: 10px; padding: 12px; text-align: center; margin-bottom: 15px; }
+.scroll-box { height: 260px; overflow-y: auto; border: 1px solid #f0f4f8; border-radius: 12px; padding: 8px; margin-bottom: 15px; }
+.char-btn { 
+    background: white !important; border: 1px solid #e2e8f0 !important; border-left: 5px solid #1299ff !important;
+    text-align: left !important; padding: 8px !important; font-size: 12px !important; transition: 0.1s !important;
+}
+.char-btn:hover { background: #f0f7ff !important; border-color: #1299ff !important; }
+.jp-btn { background: #f8fafc !important; border: 1px solid #cbd5e1 !important; color: #475569 !important; font-weight: 700 !important; border-radius: 10px !important; }
+.gen-btn { background: #1299ff !important; color: white !important; font-weight: 400 !important; border-radius: 12px !important; border: none !important; font-size: 14px !important; }
+.credit-footer { margin-top: 25px; padding: 15px; background: white; border-radius: 12px; text-align: center; border: 1px solid #eef2f6; border-bottom: 4px solid #1299ff; color: #94a3b8; font-weight: 700; font-size: 12px; letter-spacing: 2px; }
+
+/* Label Input Text background biru, teks putih */
+label[for="input_text"] {
+    background: #1299ff !important;
+    color: white !important;
+    padding: 6px 12px !important;
+    border-radius: 12px !important;
+    font-weight: 600 !important;
+    display: inline-block !important;
+    width: auto !important;
+    margin-bottom: 6px !important;
+}
+
+/* Audio player di tengah */
+.gr-audio {
+    display: flex !important;
+    justify-content: center !important;
+    background: transparent !important;
+    border: none !important;
+    padding: 0 !important;
+    margin: 10px 0 !important;
+}
+.gr-audio > div {
+    width: auto !important;
+    display: flex;
+    justify-content: center;
+}
+audio {
+    margin: 0 auto !important;
+    display: block !important;
+    width: 100% !important;
+    max-width: 360px;
+    border-radius: 30px;
+}
+label[for="audio_out"] {
+    display: none !important;
+}
+.audio-wrap {
+    background: transparent !important;
+    border: none !important;
+    box-shadow: none !important;
+}
+"""
+
+with gr.Blocks(css=css) as app:
+    with gr.Column(elem_classes="slim-card"):
+        gr.HTML("""
+            <div class="ba-header">
+                <h1>BanG Dream VITS</h1>
+                <p>🎸 Ave Mujica & MyGO 🎸</p>
+            </div>
+            <div class="status-online">
+                <div style="font-weight:800; color:#1299ff; font-size:11px; margin-bottom:4px;"><span class="dot"></span> SYSTEM STATUS</div>
+                <div style="font-size: 11px; color: #4a5568; line-height: 1.6;">
+                    Model: ✅ LOADED<br>
+                    Status: ONLINE
+                </div>
+            </div>
+        """)
+        
+        char_img = gr.Image(value=f"image/{SORTED_SPK[0]}.png", show_label=False, interactive=False, height=300, elem_classes="char-img")
+        active_spk = gr.Textbox(value=SORTED_SPK[0], visible=False)
+
+        gr.HTML("<p style='font-weight:800; font-size:11px; color:#8a99af; margin-bottom:8px;'>SELECT CHARACTER</p>")
+        
+        gr.HTML("""
+            <div class="warning-minna">
+                <div style="color:#ff9800; font-weight:800; font-size:12px; margin-bottom:4px;">🔖 PERINGATAN MINNA 🔖</div>
+                <p style="color:#666; font-size:11px; margin:0; line-height:1.4;">Setelah di pilih characternya akan Loading otomatis dan ke input, baru bisa di gunakan.</p>
+            </div>
+        """)
+
+        with gr.Column(elem_classes="scroll-box"):
+            for i in range(0, len(SORTED_SPK), 2):
+                with gr.Row():
+                    for j in range(2):
+                        if i + j < len(SORTED_SPK):
+                            name = SORTED_SPK[i+j]
+                            romaji = NAME_MAP.get(name, "Character")
+                            btn = gr.Button(f"{name} | {romaji}", elem_classes="char-btn")
+                            btn.click(lambda n=name: n, outputs=[active_spk])
+
+        input_text = gr.TextArea(label="Input Text (Japanese)", placeholder="Masukkan teks...", lines=3)
+        btn_random_input = gr.Button("🎲 Klik Acak Inputs", elem_classes="jp-btn")
+
+        with gr.Row():
+            speed_sl = gr.Slider(label="Speed", minimum=0.5, maximum=2.0, value=1.0, step=0.1)
+            sdp_sl = gr.Slider(label="SDP", minimum=0, maximum=1.0, value=0.2, step=0.1)
+
+        btn_gen = gr.Button("🎐GENERATE VOICE🎐", variant="primary", elem_classes="gen-btn")
+        audio_out = gr.Audio(label="Voice Output", type="numpy")
+        status_msg = gr.Textbox(label="Status Log", interactive=False)
+
+        gr.HTML("""<div class="credit-footer">🍃 CREATED BY MUTSUMI 🍃</div>""")
+
+    def update_image(name):
+        return f"image/{name}.png"
+    
+    def get_random_jp_text():
+        texts = [
+            "こんにちは、お会いできて光栄です。",
+            "私の音楽を聴いてください。",
+            "一生、離さないから。",
+            "私たちは、まだ終わっていない。",
+            "Ave Mujicaの世界へようこそ。",
+            "あなたは私の運命の人ですか？",
+            "心の叫びを歌に乗せて。",
+            "静かな夜に、私の声が響く。"
+        ]
+        return random.choice(texts)
+
+    active_spk.change(update_image, inputs=[active_spk], outputs=[char_img])
+    btn_random_input.click(get_random_jp_text, outputs=[input_text])
+    btn_gen.click(infer, inputs=[input_text, sdp_sl, gr.State(0.6), gr.State(0.8), speed_sl, active_spk],
+                  outputs=[status_msg, audio_out])
+
+# Launch untuk Hugging Face Spaces
+app.launch(server_name="0.0.0.0", share=False, debug=False, quiet=True)

BanG-Dream-VITS/attentions.py

@@ -0,0 +1,464 @@
+import math
+import torch
+from torch import nn
+from torch.nn import functional as F
+
+import commons
+import logging
+
+logger = logging.getLogger(__name__)
+
+
+class LayerNorm(nn.Module):
+    def __init__(self, channels, eps=1e-5):
+        super().__init__()
+        self.channels = channels
+        self.eps = eps
+
+        self.gamma = nn.Parameter(torch.ones(channels))
+        self.beta = nn.Parameter(torch.zeros(channels))
+
+    def forward(self, x):
+        x = x.transpose(1, -1)
+        x = F.layer_norm(x, (self.channels,), self.gamma, self.beta, self.eps)
+        return x.transpose(1, -1)
+
+
+@torch.jit.script
+def fused_add_tanh_sigmoid_multiply(input_a, input_b, n_channels):
+    n_channels_int = n_channels[0]
+    in_act = input_a + input_b
+    t_act = torch.tanh(in_act[:, :n_channels_int, :])
+    s_act = torch.sigmoid(in_act[:, n_channels_int:, :])
+    acts = t_act * s_act
+    return acts
+
+
+class Encoder(nn.Module):
+    def __init__(
+        self,
+        hidden_channels,
+        filter_channels,
+        n_heads,
+        n_layers,
+        kernel_size=1,
+        p_dropout=0.0,
+        window_size=4,
+        isflow=True,
+        **kwargs
+    ):
+        super().__init__()
+        self.hidden_channels = hidden_channels
+        self.filter_channels = filter_channels
+        self.n_heads = n_heads
+        self.n_layers = n_layers
+        self.kernel_size = kernel_size
+        self.p_dropout = p_dropout
+        self.window_size = window_size
+        # if isflow:
+        #  cond_layer = torch.nn.Conv1d(256, 2*hidden_channels*n_layers, 1)
+        #  self.cond_pre = torch.nn.Conv1d(hidden_channels, 2*hidden_channels, 1)
+        #  self.cond_layer = weight_norm(cond_layer, name='weight')
+        #  self.gin_channels = 256
+        self.cond_layer_idx = self.n_layers
+        if "gin_channels" in kwargs:
+            self.gin_channels = kwargs["gin_channels"]
+            if self.gin_channels != 0:
+                self.spk_emb_linear = nn.Linear(self.gin_channels, self.hidden_channels)
+                # vits2 says 3rd block, so idx is 2 by default
+                self.cond_layer_idx = (
+                    kwargs["cond_layer_idx"] if "cond_layer_idx" in kwargs else 2
+                )
+                logging.debug(self.gin_channels, self.cond_layer_idx)
+                assert (
+                    self.cond_layer_idx < self.n_layers
+                ), "cond_layer_idx should be less than n_layers"
+        self.drop = nn.Dropout(p_dropout)
+        self.attn_layers = nn.ModuleList()
+        self.norm_layers_1 = nn.ModuleList()
+        self.ffn_layers = nn.ModuleList()
+        self.norm_layers_2 = nn.ModuleList()
+        for i in range(self.n_layers):
+            self.attn_layers.append(
+                MultiHeadAttention(
+                    hidden_channels,
+                    hidden_channels,
+                    n_heads,
+                    p_dropout=p_dropout,
+                    window_size=window_size,
+                )
+            )
+            self.norm_layers_1.append(LayerNorm(hidden_channels))
+            self.ffn_layers.append(
+                FFN(
+                    hidden_channels,
+                    hidden_channels,
+                    filter_channels,
+                    kernel_size,
+                    p_dropout=p_dropout,
+                )
+            )
+            self.norm_layers_2.append(LayerNorm(hidden_channels))
+
+    def forward(self, x, x_mask, g=None):
+        attn_mask = x_mask.unsqueeze(2) * x_mask.unsqueeze(-1)
+        x = x * x_mask
+        for i in range(self.n_layers):
+            if i == self.cond_layer_idx and g is not None:
+                g = self.spk_emb_linear(g.transpose(1, 2))
+                g = g.transpose(1, 2)
+                x = x + g
+                x = x * x_mask
+            y = self.attn_layers[i](x, x, attn_mask)
+            y = self.drop(y)
+            x = self.norm_layers_1[i](x + y)
+
+            y = self.ffn_layers[i](x, x_mask)
+            y = self.drop(y)
+            x = self.norm_layers_2[i](x + y)
+        x = x * x_mask
+        return x
+
+
+class Decoder(nn.Module):
+    def __init__(
+        self,
+        hidden_channels,
+        filter_channels,
+        n_heads,
+        n_layers,
+        kernel_size=1,
+        p_dropout=0.0,
+        proximal_bias=False,
+        proximal_init=True,
+        **kwargs
+    ):
+        super().__init__()
+        self.hidden_channels = hidden_channels
+        self.filter_channels = filter_channels
+        self.n_heads = n_heads
+        self.n_layers = n_layers
+        self.kernel_size = kernel_size
+        self.p_dropout = p_dropout
+        self.proximal_bias = proximal_bias
+        self.proximal_init = proximal_init
+
+        self.drop = nn.Dropout(p_dropout)
+        self.self_attn_layers = nn.ModuleList()
+        self.norm_layers_0 = nn.ModuleList()
+        self.encdec_attn_layers = nn.ModuleList()
+        self.norm_layers_1 = nn.ModuleList()
+        self.ffn_layers = nn.ModuleList()
+        self.norm_layers_2 = nn.ModuleList()
+        for i in range(self.n_layers):
+            self.self_attn_layers.append(
+                MultiHeadAttention(
+                    hidden_channels,
+                    hidden_channels,
+                    n_heads,
+                    p_dropout=p_dropout,
+                    proximal_bias=proximal_bias,
+                    proximal_init=proximal_init,
+                )
+            )
+            self.norm_layers_0.append(LayerNorm(hidden_channels))
+            self.encdec_attn_layers.append(
+                MultiHeadAttention(
+                    hidden_channels, hidden_channels, n_heads, p_dropout=p_dropout
+                )
+            )
+            self.norm_layers_1.append(LayerNorm(hidden_channels))
+            self.ffn_layers.append(
+                FFN(
+                    hidden_channels,
+                    hidden_channels,
+                    filter_channels,
+                    kernel_size,
+                    p_dropout=p_dropout,
+                    causal=True,
+                )
+            )
+            self.norm_layers_2.append(LayerNorm(hidden_channels))
+
+    def forward(self, x, x_mask, h, h_mask):
+        """
+        x: decoder input
+        h: encoder output
+        """
+        self_attn_mask = commons.subsequent_mask(x_mask.size(2)).to(
+            device=x.device, dtype=x.dtype
+        )
+        encdec_attn_mask = h_mask.unsqueeze(2) * x_mask.unsqueeze(-1)
+        x = x * x_mask
+        for i in range(self.n_layers):
+            y = self.self_attn_layers[i](x, x, self_attn_mask)
+            y = self.drop(y)
+            x = self.norm_layers_0[i](x + y)
+
+            y = self.encdec_attn_layers[i](x, h, encdec_attn_mask)
+            y = self.drop(y)
+            x = self.norm_layers_1[i](x + y)
+
+            y = self.ffn_layers[i](x, x_mask)
+            y = self.drop(y)
+            x = self.norm_layers_2[i](x + y)
+        x = x * x_mask
+        return x
+
+
+class MultiHeadAttention(nn.Module):
+    def __init__(
+        self,
+        channels,
+        out_channels,
+        n_heads,
+        p_dropout=0.0,
+        window_size=None,
+        heads_share=True,
+        block_length=None,
+        proximal_bias=False,
+        proximal_init=False,
+    ):
+        super().__init__()
+        assert channels % n_heads == 0
+
+        self.channels = channels
+        self.out_channels = out_channels
+        self.n_heads = n_heads
+        self.p_dropout = p_dropout
+        self.window_size = window_size
+        self.heads_share = heads_share
+        self.block_length = block_length
+        self.proximal_bias = proximal_bias
+        self.proximal_init = proximal_init
+        self.attn = None
+
+        self.k_channels = channels // n_heads
+        self.conv_q = nn.Conv1d(channels, channels, 1)
+        self.conv_k = nn.Conv1d(channels, channels, 1)
+        self.conv_v = nn.Conv1d(channels, channels, 1)
+        self.conv_o = nn.Conv1d(channels, out_channels, 1)
+        self.drop = nn.Dropout(p_dropout)
+
+        if window_size is not None:
+            n_heads_rel = 1 if heads_share else n_heads
+            rel_stddev = self.k_channels**-0.5
+            self.emb_rel_k = nn.Parameter(
+                torch.randn(n_heads_rel, window_size * 2 + 1, self.k_channels)
+                * rel_stddev
+            )
+            self.emb_rel_v = nn.Parameter(
+                torch.randn(n_heads_rel, window_size * 2 + 1, self.k_channels)
+                * rel_stddev
+            )
+
+        nn.init.xavier_uniform_(self.conv_q.weight)
+        nn.init.xavier_uniform_(self.conv_k.weight)
+        nn.init.xavier_uniform_(self.conv_v.weight)
+        if proximal_init:
+            with torch.no_grad():
+                self.conv_k.weight.copy_(self.conv_q.weight)
+                self.conv_k.bias.copy_(self.conv_q.bias)
+
+    def forward(self, x, c, attn_mask=None):
+        q = self.conv_q(x)
+        k = self.conv_k(c)
+        v = self.conv_v(c)
+
+        x, self.attn = self.attention(q, k, v, mask=attn_mask)
+
+        x = self.conv_o(x)
+        return x
+
+    def attention(self, query, key, value, mask=None):
+        # reshape [b, d, t] -> [b, n_h, t, d_k]
+        b, d, t_s, t_t = (*key.size(), query.size(2))
+        query = query.view(b, self.n_heads, self.k_channels, t_t).transpose(2, 3)
+        key = key.view(b, self.n_heads, self.k_channels, t_s).transpose(2, 3)
+        value = value.view(b, self.n_heads, self.k_channels, t_s).transpose(2, 3)
+
+        scores = torch.matmul(query / math.sqrt(self.k_channels), key.transpose(-2, -1))
+        if self.window_size is not None:
+            assert (
+                t_s == t_t
+            ), "Relative attention is only available for self-attention."
+            key_relative_embeddings = self._get_relative_embeddings(self.emb_rel_k, t_s)
+            rel_logits = self._matmul_with_relative_keys(
+                query / math.sqrt(self.k_channels), key_relative_embeddings
+            )
+            scores_local = self._relative_position_to_absolute_position(rel_logits)
+            scores = scores + scores_local
+        if self.proximal_bias:
+            assert t_s == t_t, "Proximal bias is only available for self-attention."
+            scores = scores + self._attention_bias_proximal(t_s).to(
+                device=scores.device, dtype=scores.dtype
+            )
+        if mask is not None:
+            scores = scores.masked_fill(mask == 0, -1e4)
+            if self.block_length is not None:
+                assert (
+                    t_s == t_t
+                ), "Local attention is only available for self-attention."
+                block_mask = (
+                    torch.ones_like(scores)
+                    .triu(-self.block_length)
+                    .tril(self.block_length)
+                )
+                scores = scores.masked_fill(block_mask == 0, -1e4)
+        p_attn = F.softmax(scores, dim=-1)  # [b, n_h, t_t, t_s]
+        p_attn = self.drop(p_attn)
+        output = torch.matmul(p_attn, value)
+        if self.window_size is not None:
+            relative_weights = self._absolute_position_to_relative_position(p_attn)
+            value_relative_embeddings = self._get_relative_embeddings(
+                self.emb_rel_v, t_s
+            )
+            output = output + self._matmul_with_relative_values(
+                relative_weights, value_relative_embeddings
+            )
+        output = (
+            output.transpose(2, 3).contiguous().view(b, d, t_t)
+        )  # [b, n_h, t_t, d_k] -> [b, d, t_t]
+        return output, p_attn
+
+    def _matmul_with_relative_values(self, x, y):
+        """
+        x: [b, h, l, m]
+        y: [h or 1, m, d]
+        ret: [b, h, l, d]
+        """
+        ret = torch.matmul(x, y.unsqueeze(0))
+        return ret
+
+    def _matmul_with_relative_keys(self, x, y):
+        """
+        x: [b, h, l, d]
+        y: [h or 1, m, d]
+        ret: [b, h, l, m]
+        """
+        ret = torch.matmul(x, y.unsqueeze(0).transpose(-2, -1))
+        return ret
+
+    def _get_relative_embeddings(self, relative_embeddings, length):
+        2 * self.window_size + 1
+        # Pad first before slice to avoid using cond ops.
+        pad_length = max(length - (self.window_size + 1), 0)
+        slice_start_position = max((self.window_size + 1) - length, 0)
+        slice_end_position = slice_start_position + 2 * length - 1
+        if pad_length > 0:
+            padded_relative_embeddings = F.pad(
+                relative_embeddings,
+                commons.convert_pad_shape([[0, 0], [pad_length, pad_length], [0, 0]]),
+            )
+        else:
+            padded_relative_embeddings = relative_embeddings
+        used_relative_embeddings = padded_relative_embeddings[
+            :, slice_start_position:slice_end_position
+        ]
+        return used_relative_embeddings
+
+    def _relative_position_to_absolute_position(self, x):
+        """
+        x: [b, h, l, 2*l-1]
+        ret: [b, h, l, l]
+        """
+        batch, heads, length, _ = x.size()
+        # Concat columns of pad to shift from relative to absolute indexing.
+        x = F.pad(x, commons.convert_pad_shape([[0, 0], [0, 0], [0, 0], [0, 1]]))
+
+        # Concat extra elements so to add up to shape (len+1, 2*len-1).
+        x_flat = x.view([batch, heads, length * 2 * length])
+        x_flat = F.pad(
+            x_flat, commons.convert_pad_shape([[0, 0], [0, 0], [0, length - 1]])
+        )
+
+        # Reshape and slice out the padded elements.
+        x_final = x_flat.view([batch, heads, length + 1, 2 * length - 1])[
+            :, :, :length, length - 1 :
+        ]
+        return x_final
+
+    def _absolute_position_to_relative_position(self, x):
+        """
+        x: [b, h, l, l]
+        ret: [b, h, l, 2*l-1]
+        """
+        batch, heads, length, _ = x.size()
+        # pad along column
+        x = F.pad(
+            x, commons.convert_pad_shape([[0, 0], [0, 0], [0, 0], [0, length - 1]])
+        )
+        x_flat = x.view([batch, heads, length**2 + length * (length - 1)])
+        # add 0's in the beginning that will skew the elements after reshape
+        x_flat = F.pad(x_flat, commons.convert_pad_shape([[0, 0], [0, 0], [length, 0]]))
+        x_final = x_flat.view([batch, heads, length, 2 * length])[:, :, :, 1:]
+        return x_final
+
+    def _attention_bias_proximal(self, length):
+        """Bias for self-attention to encourage attention to close positions.
+        Args:
+          length: an integer scalar.
+        Returns:
+          a Tensor with shape [1, 1, length, length]
+        """
+        r = torch.arange(length, dtype=torch.float32)
+        diff = torch.unsqueeze(r, 0) - torch.unsqueeze(r, 1)
+        return torch.unsqueeze(torch.unsqueeze(-torch.log1p(torch.abs(diff)), 0), 0)
+
+
+class FFN(nn.Module):
+    def __init__(
+        self,
+        in_channels,
+        out_channels,
+        filter_channels,
+        kernel_size,
+        p_dropout=0.0,
+        activation=None,
+        causal=False,
+    ):
+        super().__init__()
+        self.in_channels = in_channels
+        self.out_channels = out_channels
+        self.filter_channels = filter_channels
+        self.kernel_size = kernel_size
+        self.p_dropout = p_dropout
+        self.activation = activation
+        self.causal = causal
+
+        if causal:
+            self.padding = self._causal_padding
+        else:
+            self.padding = self._same_padding
+
+        self.conv_1 = nn.Conv1d(in_channels, filter_channels, kernel_size)
+        self.conv_2 = nn.Conv1d(filter_channels, out_channels, kernel_size)
+        self.drop = nn.Dropout(p_dropout)
+
+    def forward(self, x, x_mask):
+        x = self.conv_1(self.padding(x * x_mask))
+        if self.activation == "gelu":
+            x = x * torch.sigmoid(1.702 * x)
+        else:
+            x = torch.relu(x)
+        x = self.drop(x)
+        x = self.conv_2(self.padding(x * x_mask))
+        return x * x_mask
+
+    def _causal_padding(self, x):
+        if self.kernel_size == 1:
+            return x
+        pad_l = self.kernel_size - 1
+        pad_r = 0
+        padding = [[0, 0], [0, 0], [pad_l, pad_r]]
+        x = F.pad(x, commons.convert_pad_shape(padding))
+        return x
+
+    def _same_padding(self, x):
+        if self.kernel_size == 1:
+            return x
+        pad_l = (self.kernel_size - 1) // 2
+        pad_r = self.kernel_size // 2
+        padding = [[0, 0], [0, 0], [pad_l, pad_r]]
+        x = F.pad(x, commons.convert_pad_shape(padding))
+        return x

BanG-Dream-VITS/attentions_onnx.py

@@ -0,0 +1,378 @@
+import math
+import torch
+from torch import nn
+from torch.nn import functional as F
+
+import commons
+import logging
+
+logger = logging.getLogger(__name__)
+
+
+class LayerNorm(nn.Module):
+    def __init__(self, channels, eps=1e-5):
+        super().__init__()
+        self.channels = channels
+        self.eps = eps
+
+        self.gamma = nn.Parameter(torch.ones(channels))
+        self.beta = nn.Parameter(torch.zeros(channels))
+
+    def forward(self, x):
+        x = x.transpose(1, -1)
+        x = F.layer_norm(x, (self.channels,), self.gamma, self.beta, self.eps)
+        return x.transpose(1, -1)
+
+
+@torch.jit.script
+def fused_add_tanh_sigmoid_multiply(input_a, input_b, n_channels):
+    n_channels_int = n_channels[0]
+    in_act = input_a + input_b
+    t_act = torch.tanh(in_act[:, :n_channels_int, :])
+    s_act = torch.sigmoid(in_act[:, n_channels_int:, :])
+    acts = t_act * s_act
+    return acts
+
+
+class Encoder(nn.Module):
+    def __init__(
+        self,
+        hidden_channels,
+        filter_channels,
+        n_heads,
+        n_layers,
+        kernel_size=1,
+        p_dropout=0.0,
+        window_size=4,
+        isflow=True,
+        **kwargs
+    ):
+        super().__init__()
+        self.hidden_channels = hidden_channels
+        self.filter_channels = filter_channels
+        self.n_heads = n_heads
+        self.n_layers = n_layers
+        self.kernel_size = kernel_size
+        self.p_dropout = p_dropout
+        self.window_size = window_size
+        # if isflow:
+        #  cond_layer = torch.nn.Conv1d(256, 2*hidden_channels*n_layers, 1)
+        #  self.cond_pre = torch.nn.Conv1d(hidden_channels, 2*hidden_channels, 1)
+        #  self.cond_layer = weight_norm(cond_layer, name='weight')
+        #  self.gin_channels = 256
+        self.cond_layer_idx = self.n_layers
+        if "gin_channels" in kwargs:
+            self.gin_channels = kwargs["gin_channels"]
+            if self.gin_channels != 0:
+                self.spk_emb_linear = nn.Linear(self.gin_channels, self.hidden_channels)
+                # vits2 says 3rd block, so idx is 2 by default
+                self.cond_layer_idx = (
+                    kwargs["cond_layer_idx"] if "cond_layer_idx" in kwargs else 2
+                )
+                logging.debug(self.gin_channels, self.cond_layer_idx)
+                assert (
+                    self.cond_layer_idx < self.n_layers
+                ), "cond_layer_idx should be less than n_layers"
+        self.drop = nn.Dropout(p_dropout)
+        self.attn_layers = nn.ModuleList()
+        self.norm_layers_1 = nn.ModuleList()
+        self.ffn_layers = nn.ModuleList()
+        self.norm_layers_2 = nn.ModuleList()
+        for i in range(self.n_layers):
+            self.attn_layers.append(
+                MultiHeadAttention(
+                    hidden_channels,
+                    hidden_channels,
+                    n_heads,
+                    p_dropout=p_dropout,
+                    window_size=window_size,
+                )
+            )
+            self.norm_layers_1.append(LayerNorm(hidden_channels))
+            self.ffn_layers.append(
+                FFN(
+                    hidden_channels,
+                    hidden_channels,
+                    filter_channels,
+                    kernel_size,
+                    p_dropout=p_dropout,
+                )
+            )
+            self.norm_layers_2.append(LayerNorm(hidden_channels))
+
+    def forward(self, x, x_mask, g=None):
+        attn_mask = x_mask.unsqueeze(2) * x_mask.unsqueeze(-1)
+        x = x * x_mask
+        for i in range(self.n_layers):
+            if i == self.cond_layer_idx and g is not None:
+                g = self.spk_emb_linear(g.transpose(1, 2))
+                g = g.transpose(1, 2)
+                x = x + g
+                x = x * x_mask
+            y = self.attn_layers[i](x, x, attn_mask)
+            y = self.drop(y)
+            x = self.norm_layers_1[i](x + y)
+
+            y = self.ffn_layers[i](x, x_mask)
+            y = self.drop(y)
+            x = self.norm_layers_2[i](x + y)
+        x = x * x_mask
+        return x
+
+
+class MultiHeadAttention(nn.Module):
+    def __init__(
+        self,
+        channels,
+        out_channels,
+        n_heads,
+        p_dropout=0.0,
+        window_size=None,
+        heads_share=True,
+        block_length=None,
+        proximal_bias=False,
+        proximal_init=False,
+    ):
+        super().__init__()
+        assert channels % n_heads == 0
+
+        self.channels = channels
+        self.out_channels = out_channels
+        self.n_heads = n_heads
+        self.p_dropout = p_dropout
+        self.window_size = window_size
+        self.heads_share = heads_share
+        self.block_length = block_length
+        self.proximal_bias = proximal_bias
+        self.proximal_init = proximal_init
+        self.attn = None
+
+        self.k_channels = channels // n_heads
+        self.conv_q = nn.Conv1d(channels, channels, 1)
+        self.conv_k = nn.Conv1d(channels, channels, 1)
+        self.conv_v = nn.Conv1d(channels, channels, 1)
+        self.conv_o = nn.Conv1d(channels, out_channels, 1)
+        self.drop = nn.Dropout(p_dropout)
+
+        if window_size is not None:
+            n_heads_rel = 1 if heads_share else n_heads
+            rel_stddev = self.k_channels**-0.5
+            self.emb_rel_k = nn.Parameter(
+                torch.randn(n_heads_rel, window_size * 2 + 1, self.k_channels)
+                * rel_stddev
+            )
+            self.emb_rel_v = nn.Parameter(
+                torch.randn(n_heads_rel, window_size * 2 + 1, self.k_channels)
+                * rel_stddev
+            )
+
+        nn.init.xavier_uniform_(self.conv_q.weight)
+        nn.init.xavier_uniform_(self.conv_k.weight)
+        nn.init.xavier_uniform_(self.conv_v.weight)
+        if proximal_init:
+            with torch.no_grad():
+                self.conv_k.weight.copy_(self.conv_q.weight)
+                self.conv_k.bias.copy_(self.conv_q.bias)
+
+    def forward(self, x, c, attn_mask=None):
+        q = self.conv_q(x)
+        k = self.conv_k(c)
+        v = self.conv_v(c)
+
+        x, self.attn = self.attention(q, k, v, mask=attn_mask)
+
+        x = self.conv_o(x)
+        return x
+
+    def attention(self, query, key, value, mask=None):
+        # reshape [b, d, t] -> [b, n_h, t, d_k]
+        b, d, t_s, t_t = (*key.size(), query.size(2))
+        query = query.view(b, self.n_heads, self.k_channels, t_t).transpose(2, 3)
+        key = key.view(b, self.n_heads, self.k_channels, t_s).transpose(2, 3)
+        value = value.view(b, self.n_heads, self.k_channels, t_s).transpose(2, 3)
+
+        scores = torch.matmul(query / math.sqrt(self.k_channels), key.transpose(-2, -1))
+        if self.window_size is not None:
+            assert (
+                t_s == t_t
+            ), "Relative attention is only available for self-attention."
+            key_relative_embeddings = self._get_relative_embeddings(self.emb_rel_k, t_s)
+            rel_logits = self._matmul_with_relative_keys(
+                query / math.sqrt(self.k_channels), key_relative_embeddings
+            )
+            scores_local = self._relative_position_to_absolute_position(rel_logits)
+            scores = scores + scores_local
+        if self.proximal_bias:
+            assert t_s == t_t, "Proximal bias is only available for self-attention."
+            scores = scores + self._attention_bias_proximal(t_s).to(
+                device=scores.device, dtype=scores.dtype
+            )
+        if mask is not None:
+            scores = scores.masked_fill(mask == 0, -1e4)
+            if self.block_length is not None:
+                assert (
+                    t_s == t_t
+                ), "Local attention is only available for self-attention."
+                block_mask = (
+                    torch.ones_like(scores)
+                    .triu(-self.block_length)
+                    .tril(self.block_length)
+                )
+                scores = scores.masked_fill(block_mask == 0, -1e4)
+        p_attn = F.softmax(scores, dim=-1)  # [b, n_h, t_t, t_s]
+        p_attn = self.drop(p_attn)
+        output = torch.matmul(p_attn, value)
+        if self.window_size is not None:
+            relative_weights = self._absolute_position_to_relative_position(p_attn)
+            value_relative_embeddings = self._get_relative_embeddings(
+                self.emb_rel_v, t_s
+            )
+            output = output + self._matmul_with_relative_values(
+                relative_weights, value_relative_embeddings
+            )
+        output = (
+            output.transpose(2, 3).contiguous().view(b, d, t_t)
+        )  # [b, n_h, t_t, d_k] -> [b, d, t_t]
+        return output, p_attn
+
+    def _matmul_with_relative_values(self, x, y):
+        """
+        x: [b, h, l, m]
+        y: [h or 1, m, d]
+        ret: [b, h, l, d]
+        """
+        ret = torch.matmul(x, y.unsqueeze(0))
+        return ret
+
+    def _matmul_with_relative_keys(self, x, y):
+        """
+        x: [b, h, l, d]
+        y: [h or 1, m, d]
+        ret: [b, h, l, m]
+        """
+        ret = torch.matmul(x, y.unsqueeze(0).transpose(-2, -1))
+        return ret
+
+    def _get_relative_embeddings(self, relative_embeddings, length):
+        max_relative_position = 2 * self.window_size + 1
+        # Pad first before slice to avoid using cond ops.
+        pad_length = max(length - (self.window_size + 1), 0)
+        slice_start_position = max((self.window_size + 1) - length, 0)
+        slice_end_position = slice_start_position + 2 * length - 1
+        if pad_length > 0:
+            padded_relative_embeddings = F.pad(
+                relative_embeddings,
+                commons.convert_pad_shape([[0, 0], [pad_length, pad_length], [0, 0]]),
+            )
+        else:
+            padded_relative_embeddings = relative_embeddings
+        used_relative_embeddings = padded_relative_embeddings[
+            :, slice_start_position:slice_end_position
+        ]
+        return used_relative_embeddings
+
+    def _relative_position_to_absolute_position(self, x):
+        """
+        x: [b, h, l, 2*l-1]
+        ret: [b, h, l, l]
+        """
+        batch, heads, length, _ = x.size()
+        # Concat columns of pad to shift from relative to absolute indexing.
+        x = F.pad(x, commons.convert_pad_shape([[0, 0], [0, 0], [0, 0], [0, 1]]))
+
+        # Concat extra elements so to add up to shape (len+1, 2*len-1).
+        x_flat = x.view([batch, heads, length * 2 * length])
+        x_flat = F.pad(
+            x_flat, commons.convert_pad_shape([[0, 0], [0, 0], [0, length - 1]])
+        )
+
+        # Reshape and slice out the padded elements.
+        x_final = x_flat.view([batch, heads, length + 1, 2 * length - 1])[
+            :, :, :length, length - 1 :
+        ]
+        return x_final
+
+    def _absolute_position_to_relative_position(self, x):
+        """
+        x: [b, h, l, l]
+        ret: [b, h, l, 2*l-1]
+        """
+        batch, heads, length, _ = x.size()
+        # padd along column
+        x = F.pad(
+            x, commons.convert_pad_shape([[0, 0], [0, 0], [0, 0], [0, length - 1]])
+        )
+        x_flat = x.view([batch, heads, length**2 + length * (length - 1)])
+        # add 0's in the beginning that will skew the elements after reshape
+        x_flat = F.pad(x_flat, commons.convert_pad_shape([[0, 0], [0, 0], [length, 0]]))
+        x_final = x_flat.view([batch, heads, length, 2 * length])[:, :, :, 1:]
+        return x_final
+
+    def _attention_bias_proximal(self, length):
+        """Bias for self-attention to encourage attention to close positions.
+        Args:
+          length: an integer scalar.
+        Returns:
+          a Tensor with shape [1, 1, length, length]
+        """
+        r = torch.arange(length, dtype=torch.float32)
+        diff = torch.unsqueeze(r, 0) - torch.unsqueeze(r, 1)
+        return torch.unsqueeze(torch.unsqueeze(-torch.log1p(torch.abs(diff)), 0), 0)
+
+
+class FFN(nn.Module):
+    def __init__(
+        self,
+        in_channels,
+        out_channels,
+        filter_channels,
+        kernel_size,
+        p_dropout=0.0,
+        activation=None,
+        causal=False,
+    ):
+        super().__init__()
+        self.in_channels = in_channels
+        self.out_channels = out_channels
+        self.filter_channels = filter_channels
+        self.kernel_size = kernel_size
+        self.p_dropout = p_dropout
+        self.activation = activation
+        self.causal = causal
+
+        if causal:
+            self.padding = self._causal_padding
+        else:
+            self.padding = self._same_padding
+
+        self.conv_1 = nn.Conv1d(in_channels, filter_channels, kernel_size)
+        self.conv_2 = nn.Conv1d(filter_channels, out_channels, kernel_size)
+        self.drop = nn.Dropout(p_dropout)
+
+    def forward(self, x, x_mask):
+        x = self.conv_1(self.padding(x * x_mask))
+        if self.activation == "gelu":
+            x = x * torch.sigmoid(1.702 * x)
+        else:
+            x = torch.relu(x)
+        x = self.drop(x)
+        x = self.conv_2(self.padding(x * x_mask))
+        return x * x_mask
+
+    def _causal_padding(self, x):
+        if self.kernel_size == 1:
+            return x
+        pad_l = self.kernel_size - 1
+        pad_r = 0
+        padding = [[0, 0], [0, 0], [pad_l, pad_r]]
+        x = F.pad(x, commons.convert_pad_shape(padding))
+        return x
+
+    def _same_padding(self, x):
+        if self.kernel_size == 1:
+            return x
+        pad_l = (self.kernel_size - 1) // 2
+        pad_r = self.kernel_size // 2
+        padding = [[0, 0], [0, 0], [pad_l, pad_r]]
+        x = F.pad(x, commons.convert_pad_shape(padding))
+        return x

BanG-Dream-VITS/bert_gen.py

@@ -0,0 +1,81 @@
+import torch
+from multiprocessing import Pool
+import commons
+import utils
+from tqdm import tqdm
+from text import check_bert_models, cleaned_text_to_sequence, get_bert
+import argparse
+import torch.multiprocessing as mp
+from config import config
+
+
+def process_line(x):
+    line, add_blank = x
+    device = config.bert_gen_config.device
+    if config.bert_gen_config.use_multi_device:
+        rank = mp.current_process()._identity
+        rank = rank[0] if len(rank) > 0 else 0
+        if torch.cuda.is_available():
+            gpu_id = rank % torch.cuda.device_count()
+            device = torch.device(f"cuda:{gpu_id}")
+        else:
+            device = torch.device("cpu")
+    wav_path, _, language_str, text, phones, tone, word2ph = line.strip().split("|")
+    phone = phones.split(" ")
+    tone = [int(i) for i in tone.split(" ")]
+    word2ph = [int(i) for i in word2ph.split(" ")]
+    word2ph = [i for i in word2ph]
+    phone, tone, language = cleaned_text_to_sequence(phone, tone, language_str)
+
+    if add_blank:
+        phone = commons.intersperse(phone, 0)
+        tone = commons.intersperse(tone, 0)
+        language = commons.intersperse(language, 0)
+        for i in range(len(word2ph)):
+            word2ph[i] = word2ph[i] * 2
+        word2ph[0] += 1
+
+    bert_path = wav_path.replace(".WAV", ".wav").replace(".wav", ".bert.pt")
+
+    try:
+        bert = torch.load(bert_path)
+        assert bert.shape[0] == 2048
+    except Exception:
+        bert = get_bert(text, word2ph, language_str, device)
+        assert bert.shape[-1] == len(phone)
+        torch.save(bert, bert_path)
+
+
+preprocess_text_config = config.preprocess_text_config
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser()
+    parser.add_argument(
+        "-c", "--config", type=str, default=config.bert_gen_config.config_path
+    )
+    parser.add_argument(
+        "--num_processes", type=int, default=config.bert_gen_config.num_processes
+    )
+    args, _ = parser.parse_known_args()
+    config_path = args.config
+    hps = utils.get_hparams_from_file(config_path)
+    check_bert_models()
+    lines = []
+    with open(hps.data.training_files, encoding="utf-8") as f:
+        lines.extend(f.readlines())
+
+    with open(hps.data.validation_files, encoding="utf-8") as f:
+        lines.extend(f.readlines())
+    add_blank = [hps.data.add_blank] * len(lines)
+
+    if len(lines) != 0:
+        num_processes = args.num_processes
+        with Pool(processes=num_processes) as pool:
+            for _ in tqdm(
+                pool.imap_unordered(process_line, zip(lines, add_blank)),
+                total=len(lines),
+            ):
+                # 这里是缩进的代码块，表示循环体
+                pass  # 使用pass语句作为占位符
+
+    print(f"bert生成完毕!, 共有{len(lines)}个bert.pt生成!")

BanG-Dream-VITS/clap_gen.py

@@ -0,0 +1,64 @@
+import argparse
+from multiprocessing import Pool, cpu_count
+
+import torch
+import torch.multiprocessing as mp
+from tqdm import tqdm
+
+import utils
+from config import config
+from clap_wrapper import get_clap_audio_feature
+import librosa
+import os
+
+os.environ["OMP_NUM_THREADS"] = "1"
+os.environ["MKL_NUM_THREADS"] = "1"
+
+
+def process_line(line):
+    device = config.emo_gen_config.device
+    if config.emo_gen_config.use_multi_device:
+        rank = mp.current_process()._identity
+        rank = rank[0] if len(rank) > 0 else 0
+        if torch.cuda.is_available():
+            gpu_id = rank % torch.cuda.device_count()
+            device = torch.device(f"cuda:{gpu_id}")
+        else:
+            device = torch.device("cpu")
+    wav_path, _, language_str, text, phones, tone, word2ph = line.strip().split("|")
+
+    clap_path = wav_path.replace(".WAV", ".wav").replace(".wav", ".emo.npy")
+    if os.path.isfile(clap_path):
+        return
+
+    audio = librosa.load(wav_path, 48000)[0]
+    # audio = librosa.resample(audio, 44100, 48000)
+
+    clap = get_clap_audio_feature(audio, device)
+    torch.save(clap, clap_path)
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser()
+    parser.add_argument(
+        "-c", "--config", type=str, default=config.emo_gen_config.config_path
+    )
+    parser.add_argument(
+        "--num_processes", type=int, default=config.emo_gen_config.num_processes
+    )
+    args, _ = parser.parse_known_args()
+    config_path = args.config
+    hps = utils.get_hparams_from_file(config_path)
+    lines = []
+    with open(hps.data.training_files, encoding="utf-8") as f:
+        lines.extend(f.readlines())
+
+    with open(hps.data.validation_files, encoding="utf-8") as f:
+        lines.extend(f.readlines())
+    if len(lines) != 0:
+        num_processes = min(args.num_processes, cpu_count())
+        with Pool(processes=num_processes) as pool:
+            for _ in tqdm(pool.imap_unordered(process_line, lines), total=len(lines)):
+                pass
+
+    print(f"clap生成完毕!, 共有{len(lines)}个emo.pt生成!")

BanG-Dream-VITS/clap_wrapper.py

@@ -0,0 +1,49 @@
+import sys
+
+import torch
+from transformers import ClapModel, ClapProcessor
+
+from config import config
+
+models = dict()
+processor = ClapProcessor.from_pretrained("./emotional/clap-htsat-fused")
+
+
+def get_clap_audio_feature(audio_data, device=config.bert_gen_config.device):
+    if (
+        sys.platform == "darwin"
+        and torch.backends.mps.is_available()
+        and device == "cpu"
+    ):
+        device = "mps"
+    if not device:
+        device = "cuda"
+    if device not in models.keys():
+        models[device] = ClapModel.from_pretrained("./emotional/clap-htsat-fused").to(
+            device
+        )
+    with torch.no_grad():
+        inputs = processor(
+            audios=audio_data, return_tensors="pt", sampling_rate=48000
+        ).to(device)
+        emb = models[device].get_audio_features(**inputs)
+    return emb.T
+
+
+def get_clap_text_feature(text, device=config.bert_gen_config.device):
+    if (
+        sys.platform == "darwin"
+        and torch.backends.mps.is_available()
+        and device == "cpu"
+    ):
+        device = "mps"
+    if not device:
+        device = "cuda"
+    if device not in models.keys():
+        models[device] = ClapModel.from_pretrained("./emotional/clap-htsat-fused").to(
+            device
+        )
+    with torch.no_grad():
+        inputs = processor(text=text, return_tensors="pt").to(device)
+        emb = models[device].get_text_features(**inputs)
+    return emb.T

BanG-Dream-VITS/classify.py

@@ -0,0 +1,180 @@
+from sklearn.cluster import *
+import os
+import numpy as np
+from config import config
+import yaml
+import argparse
+import shutil
+
+def ensure_dir(directory):
+    if not os.path.exists(directory):
+        os.makedirs(directory)
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser()
+    parser.add_argument("-a", "--algorithm", default="k", help="choose algorithm", type=str)
+    parser.add_argument("-n", "--num_clusters", default=4, help="number of clusters", type=int)
+    parser.add_argument("-r", "--range", default=4, help="number of files in a class", type=int)
+    args = parser.parse_args()
+
+    filelist_dict = {}
+    yml_result = {}
+    base_dir = "D:/Vits2/Bert-VITS2/Data/BanGDream/filelists"
+    output_dir = "D:/Vits2/classifedSample"
+
+    with open(os.path.join(base_dir, "Mygo.list"), mode="r", encoding="utf-8") as f:
+        embs = []
+        wavnames = []
+        for line in f:
+            parts = line.strip().split("|")
+            speaker = parts[1]  # 假设 speaker 是第二个部分
+            filepath = parts[0]  # 假设 filepath 是第一个部分
+            # ... 其余部分可以根据需要使用
+
+            if speaker not in filelist_dict:
+                filelist_dict[speaker] = []
+                yml_result[speaker] = {}
+            filelist_dict[speaker].append(filepath)
+
+    for speaker in filelist_dict:
+        print("\nspeaker: " + speaker)
+
+        embs = []
+        wavnames = []
+
+        for file in filelist_dict[speaker]:
+            try:
+                embs.append(np.expand_dims(np.load(f"{os.path.splitext(file)[0]}.emo.npy"), axis=0))
+                wavnames.append(file)
+            except Exception as e:
+                print(e)
+
+        if embs:
+            n_clusters = args.num_clusters
+            x = np.concatenate(embs, axis=0)
+            x = np.squeeze(x)
+
+            if args.algorithm == "b":
+                model = Birch(n_clusters=n_clusters, threshold=0.2)
+            elif args.algorithm == "s":
+                model = SpectralClustering(n_clusters=n_clusters)
+            elif args.algorithm == "a":
+                model = AgglomerativeClustering(n_clusters=n_clusters)
+            else:
+                model = KMeans(n_clusters=n_clusters, random_state=10)
+
+            y_predict = model.fit_predict(x)
+            classes = [[] for i in range(y_predict.max() + 1)]
+
+            for idx, wavname in enumerate(wavnames):
+                classes[y_predict[idx]].append(wavname)
+
+            for i in range(y_predict.max() + 1):
+                print("类别:", i, "本类中样本数量:", len(classes[i]))
+                yml_result[speaker][f"class{i}"] = []
+                class_dir = os.path.join(output_dir, speaker, f"class{i}")
+
+                num_samples_in_class = len(classes[i])
+                for j in range(min(args.range, num_samples_in_class)):
+                    wav_file = classes[i][j]
+                    print(wav_file)
+
+                    # 复制文件到新目录
+                    ensure_dir(class_dir)
+                    shutil.copy(os.path.join(base_dir, wav_file), os.path.join(class_dir, os.path.basename(wav_file)))
+
+                    yml_result[speaker][f"class{i}"].append(wav_file)
+
+    with open(os.path.join(base_dir, "emo_clustering.yml"), "w", encoding="utf-8") as f:
+        yaml.dump(yml_result, f)
+
+'''
+from sklearn.cluster import *
+import os
+import numpy as np
+from config import config
+import yaml
+import argparse
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser()
+    parser.add_argument(
+        "-a", "--algorithm", default="s", help="choose algorithm", type=str
+    )
+    parser.add_argument(
+        "-n", "--num_clusters", default=3, help="number of clusters", type=int
+    )
+    parser.add_argument(
+        "-r", "--range", default=4, help="number of files in a class", type=int
+    )
+    args = parser.parse_args()
+    filelist_dict = {}
+    yml_result = {}
+    with open(
+        "D:/Vits2/Bert-VITS2/Data/BanGDream/filelists/Mygo.list", mode="r", encoding="utf-8"
+    ) as f:
+        embs = []
+        wavnames = []
+        for line in f:
+            speaker = line.split("|")[1]
+            if speaker not in filelist_dict:
+                filelist_dict[speaker] = []
+                yml_result[speaker] = {}
+            filelist_dict[speaker].append(line.split("|")[0])
+    #print(filelist_dict)
+
+    for speaker in filelist_dict:
+        print("\nspeaker: " + speaker)
+
+        # 清空 embs 和 wavnames 列表
+        embs = []
+        wavnames = []
+
+        for file in filelist_dict[speaker]:
+            try:
+                embs.append(
+                    np.expand_dims(
+                        np.load(f"{os.path.splitext(file)[0]}.emo.npy"), axis=0
+                    )
+                )
+                wavnames.append(os.path.basename(file))
+            except Exception as e:
+                print(e)
+
+        if embs:
+        # 聚类算法类的数量
+            n_clusters = args.num_clusters
+            x = np.concatenate(embs, axis=0)
+            x = np.squeeze(x)
+            # 聚类算法类的数量
+            n_clusters = args.num_clusters
+            if args.algorithm == "b":
+                model = Birch(n_clusters=n_clusters, threshold=0.2)
+            elif args.algorithm == "s":
+                model = SpectralClustering(n_clusters=n_clusters)
+            elif args.algorithm == "a":
+                model = AgglomerativeClustering(n_clusters=n_clusters)
+            else:
+                model = KMeans(n_clusters=n_clusters, random_state=10)
+            # 可以自行尝试各种不同的聚类算法
+            y_predict = model.fit_predict(x)
+            classes = [[] for i in range(y_predict.max() + 1)]
+
+            for idx, wavname in enumerate(wavnames):
+                classes[y_predict[idx]].append(wavname)
+
+            for i in range(y_predict.max() + 1):
+                print("类别:", i, "本类中样本数量:", len(classes[i]))
+                yml_result[speaker][f"class{i}"] = []
+
+                # 修正：确保不会尝试访问超出范围的元素
+                num_samples_in_class = len(classes[i])
+                for j in range(min(args.range, num_samples_in_class)):
+                    print(classes[i][j])
+                    yml_result[speaker][f"class{i}"].append(classes[i][j])
+    with open(
+        os.path.join('D:/Vits2/Bert-VITS2/Data/BanGDream', "emo_clustering.yml"), "w", encoding="utf-8"
+    ) as f:
+        yaml.dump(yml_result, f)
+        '''

BanG-Dream-VITS/cleaner.py

@@ -0,0 +1,31 @@
+import os
+import shutil
+import psutil
+import gc
+import time
+
+def cleanup_cache(temp_dir="temp"):
+    """Membersihkan folder temp dan paksa garbage collection RAM"""
+    # 1. Bersihkan folder sementara jika ada
+    if os.path.exists(temp_dir):
+        try:
+            shutil.rmtree(temp_dir)
+            os.makedirs(temp_dir)
+            print(f"--- Folder {temp_dir} telah dibersihkan ---")
+        except Exception as e:
+            print(f"Gagal hapus folder temp: {e}")
+
+    # 2. Paksa Python lepas memori yang tidak terpakai
+    gc.collect()
+    
+    # 3. Cek penggunaan RAM saat ini
+    ram_usage = psutil.virtual_memory().percent
+    print(f"--- Penggunaan RAM saat ini: {ram_usage}% ---")
+    
+    return ram_usage
+
+def auto_clean_threshold(threshold=85):
+    """Bersihkan jika RAM di atas 85%"""
+    if psutil.virtual_memory().percent > threshold:
+        print("RAM Hampir Penuh! Memulai pembersihan darurat...")
+        gc.collect()

BanG-Dream-VITS/commons.py

@@ -0,0 +1,158 @@
+import math
+import torch
+from torch.nn import functional as F
+
+
+def init_weights(m, mean=0.0, std=0.01):
+    classname = m.__class__.__name__
+    if classname.find("Conv") != -1:
+        m.weight.data.normal_(mean, std)
+
+
+def get_padding(kernel_size, dilation=1):
+    return int((kernel_size * dilation - dilation) / 2)
+
+
+def convert_pad_shape(pad_shape):
+    layer = pad_shape[::-1]
+    pad_shape = [item for sublist in layer for item in sublist]
+    return pad_shape
+
+
+def intersperse(lst, item):
+    result = [item] * (len(lst) * 2 + 1)
+    result[1::2] = lst
+    return result
+
+
+def kl_divergence(m_p, logs_p, m_q, logs_q):
+    """KL(P||Q)"""
+    kl = (logs_q - logs_p) - 0.5
+    kl += (
+        0.5 * (torch.exp(2.0 * logs_p) + ((m_p - m_q) ** 2)) * torch.exp(-2.0 * logs_q)
+    )
+    return kl
+
+
+def rand_gumbel(shape):
+    """Sample from the Gumbel distribution, protect from overflows."""
+    uniform_samples = torch.rand(shape) * 0.99998 + 0.00001
+    return -torch.log(-torch.log(uniform_samples))
+
+
+def rand_gumbel_like(x):
+    g = rand_gumbel(x.size()).to(dtype=x.dtype, device=x.device)
+    return g
+
+
+def slice_segments(x, ids_str, segment_size=4):
+    gather_indices = ids_str.view(x.size(0), 1, 1).repeat(
+        1, x.size(1), 1
+    ) + torch.arange(segment_size, device=x.device)
+    return torch.gather(x, 2, gather_indices)
+
+
+def rand_slice_segments(x, x_lengths=None, segment_size=4):
+    b, d, t = x.size()
+    if x_lengths is None:
+        x_lengths = t
+    ids_str_max = torch.clamp(x_lengths - segment_size + 1, min=0)
+    ids_str = (torch.rand([b], device=x.device) * ids_str_max).to(dtype=torch.long)
+    ret = slice_segments(x, ids_str, segment_size)
+    return ret, ids_str
+
+
+def get_timing_signal_1d(length, channels, min_timescale=1.0, max_timescale=1.0e4):
+    position = torch.arange(length, dtype=torch.float)
+    num_timescales = channels // 2
+    log_timescale_increment = math.log(float(max_timescale) / float(min_timescale)) / (
+        num_timescales - 1
+    )
+    inv_timescales = min_timescale * torch.exp(
+        torch.arange(num_timescales, dtype=torch.float) * -log_timescale_increment
+    )
+    scaled_time = position.unsqueeze(0) * inv_timescales.unsqueeze(1)
+    signal = torch.cat([torch.sin(scaled_time), torch.cos(scaled_time)], 0)
+    signal = F.pad(signal, [0, 0, 0, channels % 2])
+    signal = signal.view(1, channels, length)
+    return signal
+
+
+def add_timing_signal_1d(x, min_timescale=1.0, max_timescale=1.0e4):
+    b, channels, length = x.size()
+    signal = get_timing_signal_1d(length, channels, min_timescale, max_timescale)
+    return x + signal.to(dtype=x.dtype, device=x.device)
+
+
+def cat_timing_signal_1d(x, min_timescale=1.0, max_timescale=1.0e4, axis=1):
+    b, channels, length = x.size()
+    signal = get_timing_signal_1d(length, channels, min_timescale, max_timescale)
+    return torch.cat([x, signal.to(dtype=x.dtype, device=x.device)], axis)
+
+
+def subsequent_mask(length):
+    mask = torch.tril(torch.ones(length, length)).unsqueeze(0).unsqueeze(0)
+    return mask
+
+
+@torch.jit.script
+def fused_add_tanh_sigmoid_multiply(input_a, input_b, n_channels):
+    n_channels_int = n_channels[0]
+    in_act = input_a + input_b
+    t_act = torch.tanh(in_act[:, :n_channels_int, :])
+    s_act = torch.sigmoid(in_act[:, n_channels_int:, :])
+    acts = t_act * s_act
+    return acts
+
+
+def convert_pad_shape(pad_shape):
+    layer = pad_shape[::-1]
+    pad_shape = [item for sublist in layer for item in sublist]
+    return pad_shape
+
+
+def shift_1d(x):
+    x = F.pad(x, convert_pad_shape([[0, 0], [0, 0], [1, 0]]))[:, :, :-1]
+    return x
+
+
+def sequence_mask(length, max_length=None):
+    if max_length is None:
+        max_length = length.max()
+    x = torch.arange(max_length, dtype=length.dtype, device=length.device)
+    return x.unsqueeze(0) < length.unsqueeze(1)
+
+
+def generate_path(duration, mask):
+    """
+    duration: [b, 1, t_x]
+    mask: [b, 1, t_y, t_x]
+    """
+
+    b, _, t_y, t_x = mask.shape
+    cum_duration = torch.cumsum(duration, -1)
+
+    cum_duration_flat = cum_duration.view(b * t_x)
+    path = sequence_mask(cum_duration_flat, t_y).to(mask.dtype)
+    path = path.view(b, t_x, t_y)
+    path = path - F.pad(path, convert_pad_shape([[0, 0], [1, 0], [0, 0]]))[:, :-1]
+    path = path.unsqueeze(1).transpose(2, 3) * mask
+    return path
+
+
+def clip_grad_value_(parameters, clip_value, norm_type=2):
+    if isinstance(parameters, torch.Tensor):
+        parameters = [parameters]
+    parameters = list(filter(lambda p: p.grad is not None, parameters))
+    norm_type = float(norm_type)
+    if clip_value is not None:
+        clip_value = float(clip_value)
+
+    total_norm = 0
+    for p in parameters:
+        param_norm = p.grad.data.norm(norm_type)
+        total_norm += param_norm.item() ** norm_type
+        if clip_value is not None:
+            p.grad.data.clamp_(min=-clip_value, max=clip_value)
+    total_norm = total_norm ** (1.0 / norm_type)
+    return total_norm

BanG-Dream-VITS/compress_model.py

@@ -0,0 +1,89 @@
+from collections import OrderedDict
+from text.symbols import symbols
+import torch
+
+from tools.log import logger
+import utils
+from models import SynthesizerTrn
+import os
+
+
+def copyStateDict(state_dict):
+    if list(state_dict.keys())[0].startswith("module"):
+        start_idx = 1
+    else:
+        start_idx = 0
+    new_state_dict = OrderedDict()
+    for k, v in state_dict.items():
+        name = ",".join(k.split(".")[start_idx:])
+        new_state_dict[name] = v
+    return new_state_dict
+
+
+def removeOptimizer(config: str, input_model: str, ishalf: bool, output_model: str):
+    hps = utils.get_hparams_from_file(config)
+
+    net_g = SynthesizerTrn(
+        len(symbols),
+        hps.data.filter_length // 2 + 1,
+        hps.train.segment_size // hps.data.hop_length,
+        n_speakers=hps.data.n_speakers,
+        **hps.model,
+    )
+
+    optim_g = torch.optim.AdamW(
+        net_g.parameters(),
+        hps.train.learning_rate,
+        betas=hps.train.betas,
+        eps=hps.train.eps,
+    )
+
+    state_dict_g = torch.load(input_model, map_location="cpu")
+    new_dict_g = copyStateDict(state_dict_g)
+    keys = []
+    for k, v in new_dict_g["model"].items():
+        if "enc_q" in k:
+            continue  # noqa: E701
+        keys.append(k)
+
+    new_dict_g = (
+        {k: new_dict_g["model"][k].half() for k in keys}
+        if ishalf
+        else {k: new_dict_g["model"][k] for k in keys}
+    )
+
+    torch.save(
+        {
+            "model": new_dict_g,
+            "iteration": 0,
+            "optimizer": optim_g.state_dict(),
+            "learning_rate": 0.0001,
+        },
+        output_model,
+    )
+
+
+if __name__ == "__main__":
+    import argparse
+
+    parser = argparse.ArgumentParser()
+    parser.add_argument("-c", "--config", type=str, default="configs/config.json")
+    parser.add_argument("-i", "--input", type=str)
+    parser.add_argument("-o", "--output", type=str, default=None)
+    parser.add_argument(
+        "-hf", "--half", action="store_true", default=False, help="Save as FP16"
+    )
+
+    args = parser.parse_args()
+
+    output = args.output
+
+    if output is None:
+        import os.path
+
+        filename, ext = os.path.splitext(args.input)
+        half = "_half" if args.half else ""
+        output = filename + "_release" + half + ext
+
+    removeOptimizer(args.config, args.input, args.half, output)
+    logger.info(f"压缩模型成功, 输出模型: {os.path.abspath(output)}")

BanG-Dream-VITS/config.py

@@ -0,0 +1,248 @@
+"""
+@Desc: 全局配置文件读取
+"""
+import argparse
+import yaml
+from typing import Dict, List
+import os
+import shutil
+import sys
+
+
+class Resample_config:
+    """重采样配置"""
+
+    def __init__(self, in_dir: str, out_dir: str, sampling_rate: int = 44100):
+        self.sampling_rate: int = sampling_rate  # 目标采样率
+        self.in_dir: str = in_dir  # 待处理音频目录路径
+        self.out_dir: str = out_dir  # 重采样输出路径
+
+    @classmethod
+    def from_dict(cls, dataset_path: str, data: Dict[str, any]):
+        """从字典中生成实例"""
+
+        # 不检查路径是否有效，此逻辑在resample.py中处理
+        data["in_dir"] = os.path.join(dataset_path, data["in_dir"])
+        data["out_dir"] = os.path.join(dataset_path, data["out_dir"])
+
+        return cls(**data)
+
+
+class Preprocess_text_config:
+    """数据预处理配置"""
+
+    def __init__(
+        self,
+        transcription_path: str,
+        cleaned_path: str,
+        train_path: str,
+        val_path: str,
+        config_path: str,
+        val_per_lang: int = 5,
+        max_val_total: int = 10000,
+        clean: bool = True,
+    ):
+        self.transcription_path: str = transcription_path  # 原始文本文件路径，文本格式应为{wav_path}|{speaker_name}|{language}|{text}。
+        self.cleaned_path: str = cleaned_path  # 数据清洗后文本路径，可以不填。不填则将在原始文本目录生成
+        self.train_path: str = train_path  # 训练集路径，可以不填。不填则将在原始文本目录生成
+        self.val_path: str = val_path  # 验证集路径，可以不填。不填则将在原始文本目录生成
+        self.config_path: str = config_path  # 配置文件路径
+        self.val_per_lang: int = val_per_lang  # 每个speaker的验证集条数
+        self.max_val_total: int = max_val_total  # 验证集最大条数，多于的会被截断并放到训练集中
+        self.clean: bool = clean  # 是否进行数据清洗
+
+    @classmethod
+    def from_dict(cls, dataset_path: str, data: Dict[str, any]):
+        """从字典中生成实例"""
+
+        data["transcription_path"] = os.path.join(
+            dataset_path, data["transcription_path"]
+        )
+        if data["cleaned_path"] == "" or data["cleaned_path"] is None:
+            data["cleaned_path"] = None
+        else:
+            data["cleaned_path"] = os.path.join(dataset_path, data["cleaned_path"])
+        data["train_path"] = os.path.join(dataset_path, data["train_path"])
+        data["val_path"] = os.path.join(dataset_path, data["val_path"])
+        data["config_path"] = os.path.join(dataset_path, data["config_path"])
+
+        return cls(**data)
+
+
+class Bert_gen_config:
+    """bert_gen 配置"""
+
+    def __init__(
+        self,
+        config_path: str,
+        num_processes: int = 2,
+        device: str = "cuda",
+        use_multi_device: bool = False,
+    ):
+        self.config_path = config_path
+        self.num_processes = num_processes
+        self.device = device
+        self.use_multi_device = use_multi_device
+
+    @classmethod
+    def from_dict(cls, dataset_path: str, data: Dict[str, any]):
+        data["config_path"] = os.path.join(dataset_path, data["config_path"])
+
+        return cls(**data)
+
+
+class Emo_gen_config:
+    """emo_gen 配置"""
+
+    def __init__(
+        self,
+        config_path: str,
+        num_processes: int = 2,
+        device: str = "cuda",
+        use_multi_device: bool = False,
+    ):
+        self.config_path = config_path
+        self.num_processes = num_processes
+        self.device = device
+        self.use_multi_device = use_multi_device
+
+    @classmethod
+    def from_dict(cls, dataset_path: str, data: Dict[str, any]):
+        data["config_path"] = os.path.join(dataset_path, data["config_path"])
+
+        return cls(**data)
+
+
+class Train_ms_config:
+    """训练配置"""
+
+    def __init__(
+        self,
+        config_path: str,
+        env: Dict[str, any],
+        base: Dict[str, any],
+        model: str,
+        num_workers: int,
+        spec_cache: bool,
+        keep_ckpts: int,
+    ):
+        self.env = env  # 需要加载的环境变量
+        self.base = base  # 底模配置
+        self.model = model  # 训练模型存储目录，该路径为相对于dataset_path的路径，而非项目根目录
+        self.config_path = config_path  # 配置文件路径
+        self.num_workers = num_workers  # worker数量
+        self.spec_cache = spec_cache  # 是否启用spec缓存
+        self.keep_ckpts = keep_ckpts  # ckpt数量
+
+    @classmethod
+    def from_dict(cls, dataset_path: str, data: Dict[str, any]):
+        # data["model"] = os.path.join(dataset_path, data["model"])
+        data["config_path"] = os.path.join(dataset_path, data["config_path"])
+
+        return cls(**data)
+
+
+class Webui_config:
+    """webui 配置"""
+
+    def __init__(
+        self,
+        device: str,
+        model: str,
+        config_path: str,
+        language_identification_library: str,
+        port: int = 7860,
+        share: bool = False,
+        debug: bool = False,
+    ):
+        self.device: str = device
+        self.model: str = model  # 端口号
+        self.config_path: str = config_path  # 是否公开部署，对外网开放
+        self.port: int = port  # 是否开启debug模式
+        self.share: bool = share  # 模型路径
+        self.debug: bool = debug  # 配置文件路径
+        self.language_identification_library: str = (
+            language_identification_library  # 语种识别库
+        )
+
+    @classmethod
+    def from_dict(cls, dataset_path: str, data: Dict[str, any]):
+        data["config_path"] = os.path.join(dataset_path, data["config_path"])
+        data["model"] = os.path.join(dataset_path, data["model"])
+        return cls(**data)
+
+
+class Server_config:
+    def __init__(
+        self, models: List[Dict[str, any]], port: int = 5000, device: str = "cuda"
+    ):
+        self.models: List[Dict[str, any]] = models  # 需要加载的所有模型的配置
+        self.port: int = port  # 端口号
+        self.device: str = device  # 模型默认使用设备
+
+    @classmethod
+    def from_dict(cls, data: Dict[str, any]):
+        return cls(**data)
+
+
+class Translate_config:
+    """翻译api配置"""
+
+    def __init__(self, app_key: str, secret_key: str):
+        self.app_key = app_key
+        self.secret_key = secret_key
+
+    @classmethod
+    def from_dict(cls, data: Dict[str, any]):
+        return cls(**data)
+
+
+class Config:
+    def __init__(self, config_path: str):
+        if not os.path.isfile(config_path) and os.path.isfile("default_config.yml"):
+            shutil.copy(src="default_config.yml", dst=config_path)
+            print(
+                f"已根据默认配置文件default_config.yml生成配置文件{config_path}。请按该配置文件的说明进行配置后重新运行。"
+            )
+            print("如无特殊需求，请勿修改default_config.yml或备份该文件。")
+            sys.exit(0)
+        with open(file=config_path, mode="r", encoding="utf-8") as file:
+            yaml_config: Dict[str, any] = yaml.safe_load(file.read())
+            dataset_path: str = yaml_config["dataset_path"]
+            openi_token: str = yaml_config["openi_token"]
+            self.dataset_path: str = dataset_path
+            self.mirror: str = yaml_config["mirror"]
+            self.openi_token: str = openi_token
+            self.resample_config: Resample_config = Resample_config.from_dict(
+                dataset_path, yaml_config["resample"]
+            )
+            self.preprocess_text_config: Preprocess_text_config = (
+                Preprocess_text_config.from_dict(
+                    dataset_path, yaml_config["preprocess_text"]
+                )
+            )
+            self.bert_gen_config: Bert_gen_config = Bert_gen_config.from_dict(
+                dataset_path, yaml_config["bert_gen"]
+            )
+            self.emo_gen_config: Emo_gen_config = Emo_gen_config.from_dict(
+                dataset_path, yaml_config["emo_gen"]
+            )
+            self.train_ms_config: Train_ms_config = Train_ms_config.from_dict(
+                dataset_path, yaml_config["train_ms"]
+            )
+            self.webui_config: Webui_config = Webui_config.from_dict(
+                dataset_path, yaml_config["webui"]
+            )
+            self.server_config: Server_config = Server_config.from_dict(
+                yaml_config["server"]
+            )
+            self.translate_config: Translate_config = Translate_config.from_dict(
+                yaml_config["translate"]
+            )
+
+
+parser = argparse.ArgumentParser()
+# 为避免与以前的config.json起冲突，将其更名如下
+parser.add_argument("-y", "--yml_config", type=str, default="config.yml")
+args, _ = parser.parse_known_args()
+config = Config(args.yml_config)

BanG-Dream-VITS/config.yml

@@ -0,0 +1,177 @@
+# 全局配置
+# 对于希望在同一时间使用多个配置文件的情况，例如两个GPU同时跑两个训练集：通过环境变量指定配置文件，不指定则默认为./config.yml
+
+# 拟提供通用路径配置，统一存放数据，避免数据放得很乱
+# 每个数据集与其对应的模型存放至统一路径下，后续所有的路径配置均为相对于datasetPath的路径
+# 不填或者填空则路径为相对于项目根目录的路径
+dataset_path: "Data/"
+
+# 模型镜像源，默认huggingface，使用openi镜像源需指定openi_token
+mirror: ""
+openi_token: ""  # openi token
+
+# resample 音频重采样配置
+# 注意， “:” 后需要加空格
+resample:
+  # 目标重采样率
+  sampling_rate: 44100
+  # 音频文件输入路径，重采样会将该路径下所有.wav音频文件重采样
+  # 请填入相对于datasetPath的相对路径
+  in_dir: "audios/raw" # 相对于根目录的路径为 /datasetPath/in_dir
+  # 音频文件重采样后输出路径
+  out_dir: "audios/wavs"
+
+
+# preprocess_text 数据集预处理相关配置
+# 注意， “:” 后需要加空格
+preprocess_text:
+  # 原始文本文件路径，文本格式应为{wav_path}|{speaker_name}|{language}|{text}。
+  transcription_path: "filelists/你的数据集文本.list"
+  # 数据清洗后文本路径，可以不填。不填则将在原始文本目录生成
+  cleaned_path: ""
+  # 训练集路径
+  train_path: "filelists/train.list"
+  # 验证集路径
+  val_path: "filelists/val.list"
+  # 配置文件路径
+  config_path: "config.json"
+  # 每个语言的验证集条数
+  val_per_lang: 4
+  # 验证集最大条数，多于的会被截断并放到训练集中
+  max_val_total: 12
+  # 是否进行数据清洗
+  clean: true
+
+
+# bert_gen 相关配置
+# 注意， “:” 后需要加空格
+bert_gen:
+  # 训练数据集配置文件路径
+  config_path: "config.json"
+  # 并行数
+  num_processes: 4
+  # 使用设备：可选项 "cuda" 显卡推理，"cpu" cpu推理
+  # 该选项同时决定了get_bert_feature的默认设备
+  device: "cuda"
+  # 使用多卡推理
+  use_multi_device: false
+
+# emo_gen 相关配置
+# 注意， “:” 后需要加空格
+emo_gen:
+  # 训练数据集配置文件路径
+  config_path: "config.json"
+  # 并行数
+  num_processes: 4
+  # 使用设备：可选项 "cuda" 显卡推理，"cpu" cpu推理
+  device: "cuda"
+  # 使用多卡推理
+  use_multi_device: false
+
+# train 训练配置
+# 注意， “:” 后需要加空格
+train_ms:
+  env:
+    MASTER_ADDR: "localhost"
+    MASTER_PORT: 10086
+    WORLD_SIZE: 1
+    LOCAL_RANK: 0
+    RANK: 0
+    # 可以填写任意名的环境变量
+    # THE_ENV_VAR_YOU_NEED_TO_USE: "1234567"
+  # 底模设置
+  base:
+    use_base_model: false
+    repo_id: "Stardust_minus/Bert-VITS2"
+    model_image: "Bert-VITS2_2.3底模" # openi网页的模型名
+  # 训练模型存储目录：与旧版本的区别，原先数据集是存放在logs/model_name下的，现在改为统一存放在Data/你的数据集/models下
+  model: "models"
+  # 配置文件路径
+  config_path: "config.json"
+  # 训练使用的worker，不建议超过CPU核心数
+  num_workers: 16
+  # 关闭此项可以节约接近50%的磁盘空间，但是可能导致实际训练速度变慢和更高的CPU使用率。
+  spec_cache: True
+  # 保存的检查点数量，多于此数目的权重会被删除来节省空间。
+  keep_ckpts: 8
+
+
+# webui webui配置
+# 注意， “:” 后需要加空格
+webui:
+  # 推理设备
+  device: "cuda"
+  # 模型路径
+  model: "models/G_8000.pth"
+  # 配置文件路径
+  config_path: "config.json"
+  # 端口号
+  port: 7860
+  # 是否公开部署，对外网开放
+  share: false
+  # 是否开启debug模式
+  debug: false
+  # 语种识别库，可选langid, fastlid
+  language_identification_library: "langid"
+
+
+# server-fastapi配置
+# 注意， “:” 后需要加空格
+# 注意，本配置下的所有配置均为相对于根目录的路径
+server:
+  # 端口号
+  port: 5000
+  # 模型默认使用设备：但是当前并没有实现这个配置。
+  device: "cuda"
+  # 需要加载的所有模型的配置，可以填多个模型，也可以不填模型，等网页成功后手动加载模型
+  # 不加载模型的配置格式：删除默认给的两个模型配置，给models赋值 [ ]，也就是空列表。参考模型2的speakers 即 models: [ ]
+  # 注意，所有模型都必须正确配置model与config的路径，空路径会导致加载错误。
+  # 也可以不填模型，等网页加载成功后手动填写models。
+  models:
+    - # 模型的路径
+      model: ""
+      # 模型config.json的路径
+      config: ""
+      # 模型使用设备，若填写则会覆盖默认配置
+      device: "cuda"
+      # 模型默认使用的语言
+      language: "ZH"
+      # 模型人物默认参数
+      # 不必填写所有人物，不填的使用默认值
+      # 暂时不用填写，当前尚未实现按人区分配置
+      speakers:
+        - speaker: "科比"
+          sdp_ratio: 0.2
+          noise_scale: 0.6
+          noise_scale_w: 0.8
+          length_scale: 1
+        - speaker: "五条悟"
+          sdp_ratio: 0.3
+          noise_scale: 0.7
+          noise_scale_w: 0.8
+          length_scale: 0.5
+        - speaker: "安倍晋三"
+          sdp_ratio: 0.2
+          noise_scale: 0.6
+          noise_scale_w: 0.8
+          length_scale: 1.2
+    - # 模型的路径
+      model: ""
+      # 模型config.json的路径
+      config: ""
+      # 模型使用设备，若填写则会覆盖默认配置
+      device: "cpu"
+      # 模型默认使用的语言
+      language: "JP"
+      # 模型人物默认参数
+      # 不必填写所有人物，不填的使用默认值
+      speakers: [ ] # 也可以不填
+
+# 百度翻译开放平台 api配置
+# api接入文档 https://api.fanyi.baidu.com/doc/21
+# 请不要在github等网站公开分享你的app id 与 key
+translate:
+  # 你的APPID
+  "app_key": "20231117001883321"
+  # 你的密钥
+  "secret_key": "lMQbvZHeJveDceLof2wf"

BanG-Dream-VITS/configs/config.json

@@ -0,0 +1,957 @@
+{
+  "train": {
+    "log_interval": 200,
+    "eval_interval": 1000,
+    "seed": 42,
+    "epochs": 1000,
+    "learning_rate": 0.0002,
+    "betas": [
+      0.8,
+      0.99
+    ],
+    "eps": 1e-09,
+    "batch_size": 16,
+    "bf16_run": false,
+    "lr_decay": 0.99995,
+    "segment_size": 16384,
+    "init_lr_ratio": 1,
+    "warmup_epochs": 0,
+    "c_mel": 45,
+    "c_kl": 1.0,
+    "c_commit": 100,
+    "skip_optimizer": true,
+    "freeze_ZH_bert": false,
+    "freeze_JP_bert": false,
+    "freeze_EN_bert": false,
+    "freeze_emo": false
+  },
+  "data": {
+    "training_files": "filelists/train.list",
+    "validation_files": "filelists/val.list",
+    "max_wav_value": 32768.0,
+    "sampling_rate": 44100,
+    "filter_length": 2048,
+    "hop_length": 512,
+    "win_length": 2048,
+    "n_mel_channels": 128,
+    "mel_fmin": 0.0,
+    "mel_fmax": null,
+    "add_blank": true,
+    "n_speakers": 850,
+    "cleaned_text": true,
+    "spk2id": {
+      "派蒙_ZH": 0,
+      "纳西妲_ZH": 1,
+      "凯亚_ZH": 2,
+      "阿贝多_ZH": 3,
+      "温迪_ZH": 4,
+      "枫原万叶_ZH": 5,
+      "钟离_ZH": 6,
+      "荒泷一斗_ZH": 7,
+      "八重神子_ZH": 8,
+      "艾尔海森_ZH": 9,
+      "提纳里_ZH": 10,
+      "迪希雅_ZH": 11,
+      "卡维_ZH": 12,
+      "宵宫_ZH": 13,
+      "那维莱特_ZH": 14,
+      "莱依拉_ZH": 15,
+      "赛诺_ZH": 16,
+      "莫娜_ZH": 17,
+      "诺艾尔_ZH": 18,
+      "托马_ZH": 19,
+      "凝光_ZH": 20,
+      "林尼_ZH": 21,
+      "北斗_ZH": 22,
+      "柯莱_ZH": 23,
+      "神里绫华_ZH": 24,
+      "可莉_ZH": 25,
+      "芭芭拉_ZH": 26,
+      "雷电将军_ZH": 27,
+      "娜维娅_ZH": 28,
+      "芙宁娜_ZH": 29,
+      "珊瑚宫心海_ZH": 30,
+      "鹿野院平藏_ZH": 31,
+      "迪奥娜_ZH": 32,
+      "琴_ZH": 33,
+      "五郎_ZH": 34,
+      "班尼特_ZH": 35,
+      "达达利亚_ZH": 36,
+      "安柏_ZH": 37,
+      "莱欧斯利_ZH": 38,
+      "夜兰_ZH": 39,
+      "妮露_ZH": 40,
+      "辛焱_ZH": 41,
+      "丽莎_ZH": 42,
+      "珐露珊_ZH": 43,
+      "魈_ZH": 44,
+      "香菱_ZH": 45,
+      "迪卢克_ZH": 46,
+      "砂糖_ZH": 47,
+      "烟绯_ZH": 48,
+      "早柚_ZH": 49,
+      "云堇_ZH": 50,
+      "刻晴_ZH": 51,
+      "重云_ZH": 52,
+      "优菈_ZH": 53,
+      "胡桃_ZH": 54,
+      "流浪者_ZH": 55,
+      "久岐忍_ZH": 56,
+      "神里绫人_ZH": 57,
+      "甘雨_ZH": 58,
+      "戴因斯雷布_ZH": 59,
+      "菲谢尔_ZH": 60,
+      "白术_ZH": 61,
+      "行秋_ZH": 62,
+      "九条裟罗_ZH": 63,
+      "夏洛蒂_ZH": 64,
+      "雷泽_ZH": 65,
+      "申鹤_ZH": 66,
+      "荧_ZH": 67,
+      "空_ZH": 68,
+      "迪娜泽黛_ZH": 69,
+      "凯瑟琳_ZH": 70,
+      "多莉_ZH": 71,
+      "坎蒂丝_ZH": 72,
+      "琳妮特_ZH": 73,
+      "萍姥姥_ZH": 74,
+      "罗莎莉亚_ZH": 75,
+      "埃德_ZH": 76,
+      "爱贝尔_ZH": 77,
+      "伊迪娅_ZH": 78,
+      "留云借风真君_ZH": 79,
+      "绮良良_ZH": 80,
+      "陌生人_ZH": 81,
+      "七七_ZH": 82,
+      "式大将_ZH": 83,
+      "瑶瑶_ZH": 84,
+      "奥兹_ZH": 85,
+      "菲米尼_ZH": 86,
+      "米卡_ZH": 87,
+      "哲平_ZH": 88,
+      "浮游水蕈兽·元素生命_ZH": 89,
+      "大肉丸_ZH": 90,
+      "托克_ZH": 91,
+      "蒂玛乌斯_ZH": 92,
+      "昆钧_ZH": 93,
+      "欧菲妮_ZH": 94,
+      "塞琉斯_ZH": 95,
+      "仆人_ZH": 96,
+      "迈勒斯_ZH": 97,
+      "希格雯_ZH": 98,
+      "阿守_ZH": 99,
+      "拉赫曼_ZH": 100,
+      "杜拉夫_ZH": 101,
+      "伊利亚斯_ZH": 102,
+      "阿晃_ZH": 103,
+      "旁白_ZH": 104,
+      "爱德琳_ZH": 105,
+      "埃洛伊_ZH": 106,
+      "德沃沙克_ZH": 107,
+      "玛乔丽_ZH": 108,
+      "塞塔蕾_ZH": 109,
+      "柊千里_ZH": 110,
+      "海芭夏_ZH": 111,
+      "九条镰治_ZH": 112,
+      "阿娜耶_ZH": 113,
+      "笼钓瓶一心_ZH": 114,
+      "回声海螺_ZH": 115,
+      "劳维克_ZH": 116,
+      "元太_ZH": 117,
+      "阿扎尔_ZH": 118,
+      "查尔斯_ZH": 119,
+      "阿洛瓦_ZH": 120,
+      "埃勒曼_ZH": 121,
+      "纳比尔_ZH": 122,
+      "莎拉_ZH": 123,
+      "康纳_ZH": 124,
+      "博来_ZH": 125,
+      "玛塞勒_ZH": 126,
+      "阿祇_ZH": 127,
+      "博士_ZH": 128,
+      "玛格丽特_ZH": 129,
+      "迪尔菲_ZH": 130,
+      "宛烟_ZH": 131,
+      "羽生田千鹤_ZH": 132,
+      "海妮耶_ZH": 133,
+      "旅行者_ZH": 134,
+      "霍夫曼_ZH": 135,
+      "佐西摩斯_ZH": 136,
+      "鹿野奈奈_ZH": 137,
+      "舒伯特_ZH": 138,
+      "天叔_ZH": 139,
+      "艾莉丝_ZH": 140,
+      "龙二_ZH": 141,
+      "莺儿_ZH": 142,
+      "嘉良_ZH": 143,
+      "一心传名刀_ZH": 144,
+      "珊瑚_ZH": 145,
+      "言笑_ZH": 146,
+      "久利须_ZH": 147,
+      "嘉玛_ZH": 148,
+      "艾文_ZH": 149,
+      "克洛琳德_ZH": 150,
+      "丹吉尔_ZH": 151,
+      "女士_ZH": 152,
+      "白老先生_ZH": 153,
+      "天目十五_ZH": 154,
+      "老孟_ZH": 155,
+      "巴达维_ZH": 156,
+      "长生_ZH": 157,
+      "吴船长_ZH": 158,
+      "拉齐_ZH": 159,
+      "艾伯特_ZH": 160,
+      "松浦_ZH": 161,
+      "埃泽_ZH": 162,
+      "阿圆_ZH": 163,
+      "莫塞伊思_ZH": 164,
+      "阿拉夫_ZH": 165,
+      "杜吉耶_ZH": 166,
+      "石头_ZH": 167,
+      "百闻_ZH": 168,
+      "波洛_ZH": 169,
+      "斯坦利_ZH": 170,
+      "博易_ZH": 171,
+      "迈蒙_ZH": 172,
+      "掇星攫辰天君_ZH": 173,
+      "毗伽尔_ZH": 174,
+      "芙卡洛斯_ZH": 175,
+      "恶龙_ZH": 176,
+      "恕筠_ZH": 177,
+      "知易_ZH": 178,
+      "克列门特_ZH": 179,
+      "大慈树王_ZH": 180,
+      "西拉杰_ZH": 181,
+      "上杉_ZH": 182,
+      "阿尔卡米_ZH": 183,
+      "纯水精灵_ZH": 184,
+      "常九爷_ZH": 185,
+      "沙扎曼_ZH": 186,
+      "田铁嘴_ZH": 187,
+      "克罗索_ZH": 188,
+      "阿巴图伊_ZH": 189,
+      "阿佩普_ZH": 190,
+      "埃尔欣根_ZH": 191,
+      "萨赫哈蒂_ZH": 192,
+      "塔杰·拉德卡尼_ZH": 193,
+      "安西_ZH": 194,
+      "陆行岩本真蕈·元素生命_ZH": 195,
+      "派蒙_JP": 196,
+      "纳西妲_JP": 197,
+      "凯亚_JP": 198,
+      "阿贝多_JP": 199,
+      "温迪_JP": 200,
+      "枫原万叶_JP": 201,
+      "钟离_JP": 202,
+      "荒泷一斗_JP": 203,
+      "八重神子_JP": 204,
+      "艾尔海森_JP": 205,
+      "提纳里_JP": 206,
+      "迪希雅_JP": 207,
+      "卡维_JP": 208,
+      "宵宫_JP": 209,
+      "那维莱特_JP": 210,
+      "莱依拉_JP": 211,
+      "赛诺_JP": 212,
+      "莫娜_JP": 213,
+      "诺艾尔_JP": 214,
+      "托马_JP": 215,
+      "凝光_JP": 216,
+      "林尼_JP": 217,
+      "北斗_JP": 218,
+      "柯莱_JP": 219,
+      "神里绫华_JP": 220,
+      "可莉_JP": 221,
+      "芭芭拉_JP": 222,
+      "雷电将军_JP": 223,
+      "娜维娅_JP": 224,
+      "芙宁娜_JP": 225,
+      "珊瑚宫心海_JP": 226,
+      "鹿野院平藏_JP": 227,
+      "迪奥娜_JP": 228,
+      "琴_JP": 229,
+      "五郎_JP": 230,
+      "班尼特_JP": 231,
+      "达达利亚_JP": 232,
+      "安柏_JP": 233,
+      "莱欧斯利_JP": 234,
+      "夜兰_JP": 235,
+      "妮露_JP": 236,
+      "辛焱_JP": 237,
+      "丽莎_JP": 238,
+      "珐露珊_JP": 239,
+      "魈_JP": 240,
+      "香菱_JP": 241,
+      "迪卢克_JP": 242,
+      "砂糖_JP": 243,
+      "烟绯_JP": 244,
+      "早柚_JP": 245,
+      "云堇_JP": 246,
+      "刻晴_JP": 247,
+      "重云_JP": 248,
+      "优菈_JP": 249,
+      "胡桃_JP": 250,
+      "流浪者_JP": 251,
+      "久岐忍_JP": 252,
+      "神里绫人_JP": 253,
+      "甘雨_JP": 254,
+      "戴因斯雷布_JP": 255,
+      "菲谢尔_JP": 256,
+      "白术_JP": 257,
+      "行秋_JP": 258,
+      "九条裟罗_JP": 259,
+      "夏洛蒂_JP": 260,
+      "雷泽_JP": 261,
+      "申鹤_JP": 262,
+      "空_JP": 263,
+      "荧_JP": 264,
+      "迪娜泽黛_JP": 265,
+      "凯瑟琳_JP": 266,
+      "多莉_JP": 267,
+      "坎蒂丝_JP": 268,
+      "琳妮特_JP": 269,
+      "萍姥姥_JP": 270,
+      "罗莎莉亚_JP": 271,
+      "埃德_JP": 272,
+      "爱贝尔_JP": 273,
+      "伊迪娅_JP": 274,
+      "留云借风真君_JP": 275,
+      "绮良良_JP": 276,
+      "陌生人_JP": 277,
+      "七七_JP": 278,
+      "式大将_JP": 279,
+      "瑶瑶_JP": 280,
+      "奥兹_JP": 281,
+      "菲米尼_JP": 282,
+      "米卡_JP": 283,
+      "哲平_JP": 284,
+      "浮游水蕈兽·元素生命_JP": 285,
+      "大肉丸_JP": 286,
+      "托克_JP": 287,
+      "蒂玛乌斯_JP": 288,
+      "昆钧_JP": 289,
+      "欧菲妮_JP": 290,
+      "塞琉斯_JP": 291,
+      "仆人_JP": 292,
+      "迈勒斯_JP": 293,
+      "希格雯_JP": 294,
+      "阿守_JP": 295,
+      "拉赫曼_JP": 296,
+      "杜拉夫_JP": 297,
+      "伊利亚斯_JP": 298,
+      "阿晃_JP": 299,
+      "旁白_JP": 300,
+      "爱德琳_JP": 301,
+      "埃洛伊_JP": 302,
+      "德沃沙克_JP": 303,
+      "玛乔丽_JP": 304,
+      "塞塔蕾_JP": 305,
+      "柊千里_JP": 306,
+      "海芭夏_JP": 307,
+      "九条镰治_JP": 308,
+      "阿娜耶_JP": 309,
+      "笼钓瓶一心_JP": 310,
+      "回声海螺_JP": 311,
+      "劳维克_JP": 312,
+      "元太_JP": 313,
+      "阿扎尔_JP": 314,
+      "查尔斯_JP": 315,
+      "阿洛瓦_JP": 316,
+      "埃勒曼_JP": 317,
+      "纳比尔_JP": 318,
+      "莎拉_JP": 319,
+      "康纳_JP": 320,
+      "博来_JP": 321,
+      "玛塞勒_JP": 322,
+      "阿祇_JP": 323,
+      "博士_JP": 324,
+      "迪尔菲_JP": 325,
+      "玛格丽特_JP": 326,
+      "宛烟_JP": 327,
+      "羽生田千鹤_JP": 328,
+      "海妮耶_JP": 329,
+      "霍夫曼_JP": 330,
+      "旅行者_JP": 331,
+      "佐西摩斯_JP": 332,
+      "舒伯特_JP": 333,
+      "鹿野奈奈_JP": 334,
+      "天叔_JP": 335,
+      "龙二_JP": 336,
+      "艾莉丝_JP": 337,
+      "莺儿_JP": 338,
+      "嘉良_JP": 339,
+      "珊瑚_JP": 340,
+      "言笑_JP": 341,
+      "一心传名刀_JP": 342,
+      "费迪南德_JP": 343,
+      "久利须_JP": 344,
+      "嘉玛_JP": 345,
+      "艾文_JP": 346,
+      "克洛琳德_JP": 347,
+      "丹吉尔_JP": 348,
+      "天目十五_JP": 349,
+      "女士_JP": 350,
+      "老孟_JP": 351,
+      "白老先生_JP": 352,
+      "舍利夫_JP": 353,
+      "巴达维_JP": 354,
+      "拉齐_JP": 355,
+      "长生_JP": 356,
+      "吴船长_JP": 357,
+      "艾伯特_JP": 358,
+      "松浦_JP": 359,
+      "埃泽_JP": 360,
+      "阿圆_JP": 361,
+      "阿拉夫_JP": 362,
+      "莫塞伊思_JP": 363,
+      "石头_JP": 364,
+      "百闻_JP": 365,
+      "杜吉耶_JP": 366,
+      "波洛_JP": 367,
+      "掇星攫辰天君_JP": 368,
+      "迈蒙_JP": 369,
+      "博易_JP": 370,
+      "诗筠_JP": 371,
+      "斯坦利_JP": 372,
+      "毗伽尔_JP": 373,
+      "芙卡洛斯_JP": 374,
+      "恶龙_JP": 375,
+      "小仓澪_JP": 376,
+      "恕筠_JP": 377,
+      "知易_JP": 378,
+      "克列门特_JP": 379,
+      "大慈树王_JP": 380,
+      "望雅_JP": 381,
+      "黑田_JP": 382,
+      "卡莉娜_JP": 383,
+      "马姆杜_JP": 384,
+      "科林斯_JP": 385,
+      "上杉_JP": 386,
+      "西拉杰_JP": 387,
+      "菲尔戈黛特_JP": 388,
+      "一平_JP": 389,
+      "纯水精灵_JP": 390,
+      "阿尔卡米_JP": 391,
+      "老戴_JP": 392,
+      "谢赫祖拜尔_JP": 393,
+      "沙扎曼_JP": 394,
+      "田铁嘴_JP": 395,
+      "小野寺_JP": 396,
+      "百识_JP": 397,
+      "克罗索_JP": 398,
+      "莱斯格_JP": 399,
+      "芷巧_JP": 400,
+      "加藤洋平_JP": 401,
+      "阿巴图伊_JP": 402,
+      "埃尔欣根_JP": 403,
+      "斯嘉莉_JP": 404,
+      "阿佩普_JP": 405,
+      "巫女_JP": 406,
+      "卡布斯_JP": 407,
+      "洛伦佐_JP": 408,
+      "萨赫哈蒂_JP": 409,
+      "娜德瓦_JP": 410,
+      "塞德娜_JP": 411,
+      "塔杰·拉德卡尼_JP": 412,
+      "绘星_JP": 413,
+      "泽田_JP": 414,
+      "安西_JP": 415,
+      "拉伊德_JP": 416,
+      "亚卡巴_JP": 417,
+      "有乐斋_JP": 418,
+      "莱昂_JP": 419,
+      "尤苏波夫_JP": 420,
+      "夏妮_JP": 421,
+      "埃舍尔_JP": 422,
+      "萨齐因_JP": 423,
+      "古山_JP": 424,
+      "自称渊上之物_JP": 425,
+      "丹羽_JP": 426,
+      "塞萨尔的日记_JP": 427,
+      "派蒙_EN": 428,
+      "纳西妲_EN": 429,
+      "凯亚_EN": 430,
+      "阿贝多_EN": 431,
+      "温迪_EN": 432,
+      "枫原万叶_EN": 433,
+      "钟离_EN": 434,
+      "荒泷一斗_EN": 435,
+      "八重神子_EN": 436,
+      "艾尔海森_EN": 437,
+      "提纳里_EN": 438,
+      "迪希雅_EN": 439,
+      "卡维_EN": 440,
+      "宵宫_EN": 441,
+      "莱依拉_EN": 442,
+      "那维莱特_EN": 443,
+      "赛诺_EN": 444,
+      "莫娜_EN": 445,
+      "诺艾尔_EN": 446,
+      "托马_EN": 447,
+      "凝光_EN": 448,
+      "林尼_EN": 449,
+      "北斗_EN": 450,
+      "柯莱_EN": 451,
+      "神里绫华_EN": 452,
+      "可莉_EN": 453,
+      "芭芭拉_EN": 454,
+      "雷电将军_EN": 455,
+      "娜维娅_EN": 456,
+      "芙宁娜_EN": 457,
+      "珊瑚宫心海_EN": 458,
+      "鹿野院平藏_EN": 459,
+      "迪奥娜_EN": 460,
+      "五郎_EN": 461,
+      "琴_EN": 462,
+      "班尼特_EN": 463,
+      "达达利亚_EN": 464,
+      "安柏_EN": 465,
+      "莱欧斯利_EN": 466,
+      "夜兰_EN": 467,
+      "妮露_EN": 468,
+      "辛焱_EN": 469,
+      "珐露珊_EN": 470,
+      "丽莎_EN": 471,
+      "魈_EN": 472,
+      "香菱_EN": 473,
+      "迪卢克_EN": 474,
+      "砂糖_EN": 475,
+      "烟绯_EN": 476,
+      "早柚_EN": 477,
+      "云堇_EN": 478,
+      "刻晴_EN": 479,
+      "重云_EN": 480,
+      "优菈_EN": 481,
+      "胡桃_EN": 482,
+      "流浪者_EN": 483,
+      "久岐忍_EN": 484,
+      "神里绫人_EN": 485,
+      "甘雨_EN": 486,
+      "戴因斯雷布_EN": 487,
+      "菲谢尔_EN": 488,
+      "白术_EN": 489,
+      "行秋_EN": 490,
+      "九条裟罗_EN": 491,
+      "夏洛蒂_EN": 492,
+      "雷泽_EN": 493,
+      "申鹤_EN": 494,
+      "荧_EN": 495,
+      "空_EN": 496,
+      "迪娜泽黛_EN": 497,
+      "凯瑟琳_EN": 498,
+      "多莉_EN": 499,
+      "坎蒂丝_EN": 500,
+      "琳妮特_EN": 501,
+      "萍姥姥_EN": 502,
+      "罗莎莉亚_EN": 503,
+      "埃德_EN": 504,
+      "爱贝尔_EN": 505,
+      "伊迪娅_EN": 506,
+      "留云借风真君_EN": 507,
+      "绮良良_EN": 508,
+      "陌生人_EN": 509,
+      "七七_EN": 510,
+      "式大将_EN": 511,
+      "瑶瑶_EN": 512,
+      "奥兹_EN": 513,
+      "菲米尼_EN": 514,
+      "米卡_EN": 515,
+      "哲平_EN": 516,
+      "浮游水蕈兽·元素生命_EN": 517,
+      "大肉丸_EN": 518,
+      "托克_EN": 519,
+      "蒂玛乌斯_EN": 520,
+      "昆钧_EN": 521,
+      "欧菲妮_EN": 522,
+      "塞琉斯_EN": 523,
+      "仆人_EN": 524,
+      "迈勒斯_EN": 525,
+      "希格雯_EN": 526,
+      "阿守_EN": 527,
+      "拉赫曼_EN": 528,
+      "杜拉夫_EN": 529,
+      "伊利亚斯_EN": 530,
+      "阿晃_EN": 531,
+      "旁白_EN": 532,
+      "爱德琳_EN": 533,
+      "埃洛伊_EN": 534,
+      "德沃沙克_EN": 535,
+      "玛乔丽_EN": 536,
+      "塞塔蕾_EN": 537,
+      "柊千里_EN": 538,
+      "海芭夏_EN": 539,
+      "九条镰治_EN": 540,
+      "阿娜耶_EN": 541,
+      "笼钓瓶一心_EN": 542,
+      "回声海螺_EN": 543,
+      "劳维克_EN": 544,
+      "元太_EN": 545,
+      "阿扎尔_EN": 546,
+      "查尔斯_EN": 547,
+      "阿洛瓦_EN": 548,
+      "埃勒曼_EN": 549,
+      "纳比尔_EN": 550,
+      "莎拉_EN": 551,
+      "康纳_EN": 552,
+      "博来_EN": 553,
+      "玛塞勒_EN": 554,
+      "阿祇_EN": 555,
+      "博士_EN": 556,
+      "迪尔菲_EN": 557,
+      "宛烟_EN": 558,
+      "玛格丽特_EN": 559,
+      "羽生田千鹤_EN": 560,
+      "海妮耶_EN": 561,
+      "霍夫曼_EN": 562,
+      "旅行者_EN": 563,
+      "佐西摩斯_EN": 564,
+      "鹿野奈奈_EN": 565,
+      "舒伯特_EN": 566,
+      "天叔_EN": 567,
+      "艾莉丝_EN": 568,
+      "龙二_EN": 569,
+      "莺儿_EN": 570,
+      "嘉良_EN": 571,
+      "珊瑚_EN": 572,
+      "费迪南德_EN": 573,
+      "言笑_EN": 574,
+      "一心传名刀_EN": 575,
+      "久利须_EN": 576,
+      "嘉玛_EN": 577,
+      "艾文_EN": 578,
+      "克洛琳德_EN": 579,
+      "丹吉尔_EN": 580,
+      "女士_EN": 581,
+      "天目十五_EN": 582,
+      "老孟_EN": 583,
+      "白老先生_EN": 584,
+      "舍利夫_EN": 585,
+      "巴达维_EN": 586,
+      "拉齐_EN": 587,
+      "长生_EN": 588,
+      "吴船长_EN": 589,
+      "艾伯特_EN": 590,
+      "松浦_EN": 591,
+      "埃泽_EN": 592,
+      "阿圆_EN": 593,
+      "阿拉夫_EN": 594,
+      "莫塞伊思_EN": 595,
+      "石头_EN": 596,
+      "百闻_EN": 597,
+      "杜吉耶_EN": 598,
+      "波洛_EN": 599,
+      "斯坦利_EN": 600,
+      "掇星攫辰天君_EN": 601,
+      "迈蒙_EN": 602,
+      "博易_EN": 603,
+      "诗筠_EN": 604,
+      "毗伽尔_EN": 605,
+      "慧心_EN": 606,
+      "芙卡洛斯_EN": 607,
+      "恶龙_EN": 608,
+      "小仓澪_EN": 609,
+      "恕筠_EN": 610,
+      "知易_EN": 611,
+      "克列门特_EN": 612,
+      "大慈树王_EN": 613,
+      "维多利亚_EN": 614,
+      "黑田_EN": 615,
+      "马姆杜_EN": 616,
+      "科林斯_EN": 617,
+      "上杉_EN": 618,
+      "西拉杰_EN": 619,
+      "宁禄_EN": 620,
+      "纯水精灵_EN": 621,
+      "常九爷_EN": 622,
+      "阿尔卡米_EN": 623,
+      "沙扎曼_EN": 624,
+      "田铁嘴_EN": 625,
+      "加萨尼_EN": 626,
+      "克罗索_EN": 627,
+      "星稀_EN": 628,
+      "莱斯格_EN": 629,
+      "阿巴图伊_EN": 630,
+      "埃尔欣根_EN": 631,
+      "阿佩普_EN": 632,
+      "萨赫哈蒂_EN": 633,
+      "洛伦佐_EN": 634,
+      "塔杰·拉德卡尼_EN": 635,
+      "泽田_EN": 636,
+      "安西_EN": 637,
+      "埃舍尔_EN": 638,
+      "三月七_ZH": 639,
+      "丹恒_ZH": 640,
+      "希儿_ZH": 641,
+      "娜塔莎_ZH": 642,
+      "希露瓦_ZH": 643,
+      "瓦尔特_ZH": 644,
+      "佩拉_ZH": 645,
+      "布洛妮娅_ZH": 646,
+      "虎克_ZH": 647,
+      "素裳_ZH": 648,
+      "克拉拉_ZH": 649,
+      "符玄_ZH": 650,
+      "白露_ZH": 651,
+      "杰帕德_ZH": 652,
+      "景元_ZH": 653,
+      "藿藿_ZH": 654,
+      "姬子_ZH": 655,
+      "穹_ZH": 656,
+      "星_ZH": 657,
+      "卡芙卡_ZH": 658,
+      "桂乃芬_ZH": 659,
+      "艾丝妲_ZH": 660,
+      "玲可_ZH": 661,
+      "彦卿_ZH": 662,
+      "托帕_ZH": 663,
+      "驭空_ZH": 664,
+      "浮烟_ZH": 665,
+      "停云_ZH": 666,
+      "镜流_ZH": 667,
+      "罗刹_ZH": 668,
+      "卢卡_ZH": 669,
+      "史瓦罗_ZH": 670,
+      "黑塔_ZH": 671,
+      "桑博_ZH": 672,
+      "伦纳德_ZH": 673,
+      "明曦_ZH": 674,
+      "银狼_ZH": 675,
+      "帕姆_ZH": 676,
+      "青雀_ZH": 677,
+      "乔瓦尼_ZH": 678,
+      "公输师傅_ZH": 679,
+      "晴霓_ZH": 680,
+      "螺丝咕姆_ZH": 681,
+      "阿兰_ZH": 682,
+      "奥列格_ZH": 683,
+      "丹枢_ZH": 684,
+      "尾巴_ZH": 685,
+      "寒鸦_ZH": 686,
+      "雪衣_ZH": 687,
+      "可可利亚_ZH": 688,
+      "青镞_ZH": 689,
+      "半夏_ZH": 690,
+      "银枝_ZH": 691,
+      "大毫_ZH": 692,
+      "霄翰_ZH": 693,
+      "信使_ZH": 694,
+      "费斯曼_ZH": 695,
+      "绿芙蓉_ZH": 696,
+      "金人会长_ZH": 697,
+      "维利特_ZH": 698,
+      "维尔德_ZH": 699,
+      "斯科特_ZH": 700,
+      "卡波特_ZH": 701,
+      "刃_ZH": 702,
+      "岩明_ZH": 703,
+      "浣溪_ZH": 704,
+      "三月七_JP": 705,
+      "丹恒_JP": 706,
+      "希儿_JP": 707,
+      "娜塔莎_JP": 708,
+      "希露瓦_JP": 709,
+      "瓦尔特_JP": 710,
+      "佩拉_JP": 711,
+      "布洛妮娅_JP": 712,
+      "虎克_JP": 713,
+      "素裳_JP": 714,
+      "克拉拉_JP": 715,
+      "符玄_JP": 716,
+      "白露_JP": 717,
+      "杰帕德_JP": 718,
+      "景元_JP": 719,
+      "藿藿_JP": 720,
+      "姬子_JP": 721,
+      "卡芙卡_JP": 722,
+      "穹_JP": 723,
+      "星_JP": 724,
+      "桂乃芬_JP": 725,
+      "艾丝妲_JP": 726,
+      "彦卿_JP": 727,
+      "玲可_JP": 728,
+      "托帕_JP": 729,
+      "驭空_JP": 730,
+      "浮烟_JP": 731,
+      "停云_JP": 732,
+      "镜流_JP": 733,
+      "罗刹_JP": 734,
+      "卢卡_JP": 735,
+      "史瓦罗_JP": 736,
+      "黑塔_JP": 737,
+      "桑博_JP": 738,
+      "伦纳德_JP": 739,
+      "明曦_JP": 740,
+      "银狼_JP": 741,
+      "帕姆_JP": 742,
+      "青雀_JP": 743,
+      "乔瓦尼_JP": 744,
+      "公输师傅_JP": 745,
+      "晴霓_JP": 746,
+      "螺丝咕姆_JP": 747,
+      "阿兰_JP": 748,
+      "奥列格_JP": 749,
+      "丹枢_JP": 750,
+      "尾巴_JP": 751,
+      "寒鸦_JP": 752,
+      "雪衣_JP": 753,
+      "可可利亚_JP": 754,
+      "青镞_JP": 755,
+      "半夏_JP": 756,
+      "银枝_JP": 757,
+      "大毫_JP": 758,
+      "霄翰_JP": 759,
+      "信使_JP": 760,
+      "费斯曼_JP": 761,
+      "绿芙蓉_JP": 762,
+      "金人会长_JP": 763,
+      "维利特_JP": 764,
+      "维尔德_JP": 765,
+      "斯科特_JP": 766,
+      "刃_JP": 767,
+      "卡波特_JP": 768,
+      "岩明_JP": 769,
+      "浣溪_JP": 770,
+      "净砚_JP": 771,
+      "紫月季_JP": 772,
+      "歌蒂_JP": 773,
+      "奇怪的云骑_JP": 774,
+      "幻胧_JP": 775,
+      "斯薇塔_JP": 776,
+      "隐书_JP": 777,
+      "三月七_EN": 778,
+      "丹恒_EN": 779,
+      "希儿_EN": 780,
+      "娜塔莎_EN": 781,
+      "希露瓦_EN": 782,
+      "瓦尔特_EN": 783,
+      "佩拉_EN": 784,
+      "布洛妮娅_EN": 785,
+      "虎克_EN": 786,
+      "素裳_EN": 787,
+      "克拉拉_EN": 788,
+      "符玄_EN": 789,
+      "白露_EN": 790,
+      "杰帕德_EN": 791,
+      "景元_EN": 792,
+      "藿藿_EN": 793,
+      "姬子_EN": 794,
+      "卡芙卡_EN": 795,
+      "穹_EN": 796,
+      "星_EN": 797,
+      "桂乃芬_EN": 798,
+      "艾丝妲_EN": 799,
+      "彦卿_EN": 800,
+      "玲可_EN": 801,
+      "托帕_EN": 802,
+      "驭空_EN": 803,
+      "浮烟_EN": 804,
+      "停云_EN": 805,
+      "镜流_EN": 806,
+      "罗刹_EN": 807,
+      "卢卡_EN": 808,
+      "史瓦罗_EN": 809,
+      "黑塔_EN": 810,
+      "桑博_EN": 811,
+      "伦纳德_EN": 812,
+      "明曦_EN": 813,
+      "银狼_EN": 814,
+      "帕姆_EN": 815,
+      "青雀_EN": 816,
+      "乔瓦尼_EN": 817,
+      "公输师傅_EN": 818,
+      "晴霓_EN": 819,
+      "螺丝咕姆_EN": 820,
+      "阿兰_EN": 821,
+      "奥列格_EN": 822,
+      "丹枢_EN": 823,
+      "尾巴_EN": 824,
+      "寒鸦_EN": 825,
+      "雪衣_EN": 826,
+      "可可利亚_EN": 827,
+      "青镞_EN": 828,
+      "半夏_EN": 829,
+      "银枝_EN": 830,
+      "大毫_EN": 831,
+      "霄翰_EN": 832,
+      "信使_EN": 833,
+      "费斯曼_EN": 834,
+      "绿芙蓉_EN": 835,
+      "金人会长_EN": 836,
+      "维利特_EN": 837,
+      "维尔德_EN": 838,
+      "刃_EN": 839,
+      "卡波特_EN": 840,
+      "岩明_EN": 841,
+      "浣溪_EN": 842,
+      "紫月季_EN": 843,
+      "幻胧_EN": 844,
+      "女声_EN": 845,
+      "陆景和": 846,
+      "莫弈": 847,
+      "左然": 848,
+      "夏彦": 849
+    }
+  },
+  "model": {
+    "use_spk_conditioned_encoder": true,
+    "use_noise_scaled_mas": true,
+    "use_mel_posterior_encoder": false,
+    "use_duration_discriminator": true,
+    "inter_channels": 192,
+    "hidden_channels": 192,
+    "filter_channels": 768,
+    "n_heads": 2,
+    "n_layers": 6,
+    "kernel_size": 3,
+    "p_dropout": 0.1,
+    "resblock": "1",
+    "resblock_kernel_sizes": [
+      3,
+      7,
+      11
+    ],
+    "resblock_dilation_sizes": [
+      [
+        1,
+        3,
+        5
+      ],
+      [
+        1,
+        3,
+        5
+      ],
+      [
+        1,
+        3,
+        5
+      ]
+    ],
+    "upsample_rates": [
+      8,
+      8,
+      2,
+      2,
+      2
+    ],
+    "upsample_initial_channel": 512,
+    "upsample_kernel_sizes": [
+      16,
+      16,
+      8,
+      2,
+      2
+    ],
+    "n_layers_q": 3,
+    "use_spectral_norm": false,
+    "gin_channels": 512,
+    "slm": {
+        "model": "./slm/wavlm-base-plus",
+        "sr": 16000,
+        "hidden": 768,
+        "nlayers": 13,
+        "initial_channel": 64
+    }
+  },
+  "version": "2.3"
+}

BanG-Dream-VITS/css/custom.css

@@ -0,0 +1,18 @@
+
+#yml_code {
+    height: 600px;
+    flex-grow: inherit;
+    overflow-y: auto;
+}
+
+#json_code {
+    height: 600px;
+    flex-grow: inherit;
+    overflow-y: auto;
+}
+
+#gpu_code {
+    height: 300px;
+    flex-grow: inherit;
+    overflow-y: auto;
+}

BanG-Dream-VITS/data_utils.py

@@ -0,0 +1,404 @@
+import os
+import random
+import torch
+import torch.utils.data
+from tqdm import tqdm
+from tools.log import logger
+import commons
+from mel_processing import spectrogram_torch, mel_spectrogram_torch
+from utils import load_wav_to_torch, load_filepaths_and_text
+from text import cleaned_text_to_sequence
+from config import config
+
+"""Multi speaker version"""
+
+
+class TextAudioSpeakerLoader(torch.utils.data.Dataset):
+    """
+    1) loads audio, speaker_id, text pairs
+    2) normalizes text and converts them to sequences of integers
+    3) computes spectrograms from audio files.
+    """
+
+    def __init__(self, audiopaths_sid_text, hparams):
+        self.audiopaths_sid_text = load_filepaths_and_text(audiopaths_sid_text)
+        self.max_wav_value = hparams.max_wav_value
+        self.sampling_rate = hparams.sampling_rate
+        self.filter_length = hparams.filter_length
+        self.hop_length = hparams.hop_length
+        self.win_length = hparams.win_length
+        self.sampling_rate = hparams.sampling_rate
+        self.spk_map = hparams.spk2id
+        self.hparams = hparams
+
+        self.use_mel_spec_posterior = getattr(
+            hparams, "use_mel_posterior_encoder", False
+        )
+        if self.use_mel_spec_posterior:
+            self.n_mel_channels = getattr(hparams, "n_mel_channels", 80)
+
+        self.cleaned_text = getattr(hparams, "cleaned_text", False)
+
+        self.add_blank = hparams.add_blank
+        self.min_text_len = getattr(hparams, "min_text_len", 1)
+        self.max_text_len = getattr(hparams, "max_text_len", 384)
+
+        random.seed(1234)
+        random.shuffle(self.audiopaths_sid_text)
+        self._filter()
+
+    def _filter(self):
+        """
+        Filter text & store spec lengths
+        """
+        # Store spectrogram lengths for Bucketing
+        # wav_length ~= file_size / (wav_channels * Bytes per dim) = file_size / (1 * 2)
+        # spec_length = wav_length // hop_length
+
+        audiopaths_sid_text_new = []
+        lengths = []
+        skipped = 0
+        logger.info("Init dataset...")
+        for _id, spk, language, text, phones, tone, word2ph in tqdm(
+            self.audiopaths_sid_text
+        ):
+            audiopath = f"{_id}"
+            if self.min_text_len <= len(phones) and len(phones) <= self.max_text_len:
+                phones = phones.split(" ")
+                tone = [int(i) for i in tone.split(" ")]
+                word2ph = [int(i) for i in word2ph.split(" ")]
+                audiopaths_sid_text_new.append(
+                    [audiopath, spk, language, text, phones, tone, word2ph]
+                )
+                lengths.append(os.path.getsize(audiopath) // (2 * self.hop_length))
+            else:
+                skipped += 1
+        logger.info(
+            "skipped: "
+            + str(skipped)
+            + ", total: "
+            + str(len(self.audiopaths_sid_text))
+        )
+        self.audiopaths_sid_text = audiopaths_sid_text_new
+        self.lengths = lengths
+
+    def get_audio_text_speaker_pair(self, audiopath_sid_text):
+        # separate filename, speaker_id and text
+        audiopath, sid, language, text, phones, tone, word2ph = audiopath_sid_text
+
+        bert, ja_bert, en_bert, phones, tone, language = self.get_text(
+            text, word2ph, phones, tone, language, audiopath
+        )
+
+        spec, wav = self.get_audio(audiopath)
+        sid = torch.LongTensor([int(self.spk_map[sid])])
+
+        return (phones, spec, wav, sid, tone, language, bert, ja_bert, en_bert)
+
+    def get_audio(self, filename):
+        audio, sampling_rate = load_wav_to_torch(filename)
+        if sampling_rate != self.sampling_rate:
+            raise ValueError(
+                "{} {} SR doesn't match target {} SR".format(
+                    filename, sampling_rate, self.sampling_rate
+                )
+            )
+        audio_norm = audio / self.max_wav_value
+        audio_norm = audio_norm.unsqueeze(0)
+        spec_filename = filename.replace(".wav", ".spec.pt")
+        if self.use_mel_spec_posterior:
+            spec_filename = spec_filename.replace(".spec.pt", ".mel.pt")
+        try:
+            spec = torch.load(spec_filename)
+        except:
+            if self.use_mel_spec_posterior:
+                spec = mel_spectrogram_torch(
+                    audio_norm,
+                    self.filter_length,
+                    self.n_mel_channels,
+                    self.sampling_rate,
+                    self.hop_length,
+                    self.win_length,
+                    self.hparams.mel_fmin,
+                    self.hparams.mel_fmax,
+                    center=False,
+                )
+            else:
+                spec = spectrogram_torch(
+                    audio_norm,
+                    self.filter_length,
+                    self.sampling_rate,
+                    self.hop_length,
+                    self.win_length,
+                    center=False,
+                )
+            spec = torch.squeeze(spec, 0)
+            if config.train_ms_config.spec_cache:
+                torch.save(spec, spec_filename)
+        return spec, audio_norm
+
+    def get_text(self, text, word2ph, phone, tone, language_str, wav_path):
+        phone, tone, language = cleaned_text_to_sequence(phone, tone, language_str)
+        if self.add_blank:
+            phone = commons.intersperse(phone, 0)
+            tone = commons.intersperse(tone, 0)
+            language = commons.intersperse(language, 0)
+            for i in range(len(word2ph)):
+                word2ph[i] = word2ph[i] * 2
+            word2ph[0] += 1
+        bert_path = wav_path.replace(".wav", ".bert.pt")
+        try:
+            bert_ori = torch.load(bert_path)
+            assert bert_ori.shape[-1] == len(phone)
+        except Exception as e:
+            logger.warning("Bert load Failed")
+            logger.warning(e)
+
+        if language_str == "ZH":
+            bert = bert_ori
+            ja_bert = torch.randn(1024, len(phone))
+            en_bert = torch.randn(1024, len(phone))
+        elif language_str == "JP":
+            bert = torch.randn(1024, len(phone))
+            ja_bert = bert_ori
+            en_bert = torch.randn(1024, len(phone))
+        elif language_str == "EN":
+            bert = torch.randn(1024, len(phone))
+            ja_bert = torch.randn(1024, len(phone))
+            en_bert = bert_ori
+        phone = torch.LongTensor(phone)
+        tone = torch.LongTensor(tone)
+        language = torch.LongTensor(language)
+        return bert, ja_bert, en_bert, phone, tone, language
+
+    def get_sid(self, sid):
+        sid = torch.LongTensor([int(sid)])
+        return sid
+
+    def __getitem__(self, index):
+        return self.get_audio_text_speaker_pair(self.audiopaths_sid_text[index])
+
+    def __len__(self):
+        return len(self.audiopaths_sid_text)
+
+
+class TextAudioSpeakerCollate:
+    """Zero-pads model inputs and targets"""
+
+    def __init__(self, return_ids=False):
+        self.return_ids = return_ids
+
+    def __call__(self, batch):
+        """Collate's training batch from normalized text, audio and speaker identities
+        PARAMS
+        ------
+        batch: [text_normalized, spec_normalized, wav_normalized, sid]
+        """
+        # Right zero-pad all one-hot text sequences to max input length
+        _, ids_sorted_decreasing = torch.sort(
+            torch.LongTensor([x[1].size(1) for x in batch]), dim=0, descending=True
+        )
+
+        max_text_len = max([len(x[0]) for x in batch])
+        max_spec_len = max([x[1].size(1) for x in batch])
+        max_wav_len = max([x[2].size(1) for x in batch])
+
+        text_lengths = torch.LongTensor(len(batch))
+        spec_lengths = torch.LongTensor(len(batch))
+        wav_lengths = torch.LongTensor(len(batch))
+        sid = torch.LongTensor(len(batch))
+
+        text_padded = torch.LongTensor(len(batch), max_text_len)
+        tone_padded = torch.LongTensor(len(batch), max_text_len)
+        language_padded = torch.LongTensor(len(batch), max_text_len)
+        bert_padded = torch.FloatTensor(len(batch), 1024, max_text_len)
+        ja_bert_padded = torch.FloatTensor(len(batch), 1024, max_text_len)
+        en_bert_padded = torch.FloatTensor(len(batch), 1024, max_text_len)
+
+        spec_padded = torch.FloatTensor(len(batch), batch[0][1].size(0), max_spec_len)
+        wav_padded = torch.FloatTensor(len(batch), 1, max_wav_len)
+        text_padded.zero_()
+        tone_padded.zero_()
+        language_padded.zero_()
+        spec_padded.zero_()
+        wav_padded.zero_()
+        bert_padded.zero_()
+        ja_bert_padded.zero_()
+        en_bert_padded.zero_()
+
+        for i in range(len(ids_sorted_decreasing)):
+            row = batch[ids_sorted_decreasing[i]]
+
+            text = row[0]
+            text_padded[i, : text.size(0)] = text
+            text_lengths[i] = text.size(0)
+
+            spec = row[1]
+            spec_padded[i, :, : spec.size(1)] = spec
+            spec_lengths[i] = spec.size(1)
+
+            wav = row[2]
+            wav_padded[i, :, : wav.size(1)] = wav
+            wav_lengths[i] = wav.size(1)
+
+            sid[i] = row[3]
+
+            tone = row[4]
+            tone_padded[i, : tone.size(0)] = tone
+
+            language = row[5]
+            language_padded[i, : language.size(0)] = language
+
+            bert = row[6]
+            bert_padded[i, :, : bert.size(1)] = bert
+
+            ja_bert = row[7]
+            ja_bert_padded[i, :, : ja_bert.size(1)] = ja_bert
+
+            en_bert = row[8]
+            en_bert_padded[i, :, : en_bert.size(1)] = en_bert
+
+        return (
+            text_padded,
+            text_lengths,
+            spec_padded,
+            spec_lengths,
+            wav_padded,
+            wav_lengths,
+            sid,
+            tone_padded,
+            language_padded,
+            bert_padded,
+            ja_bert_padded,
+            en_bert_padded,
+        )
+
+
+class DistributedBucketSampler(torch.utils.data.distributed.DistributedSampler):
+    """
+    Maintain similar input lengths in a batch.
+    Length groups are specified by boundaries.
+    Ex) boundaries = [b1, b2, b3] -> any batch is included either {x | b1 < length(x) <=b2} or {x | b2 < length(x) <= b3}.
+
+    It removes samples which are not included in the boundaries.
+    Ex) boundaries = [b1, b2, b3] -> any x s.t. length(x) <= b1 or length(x) > b3 are discarded.
+    """
+
+    def __init__(
+        self,
+        dataset,
+        batch_size,
+        boundaries,
+        num_replicas=None,
+        rank=None,
+        shuffle=True,
+    ):
+        super().__init__(dataset, num_replicas=num_replicas, rank=rank, shuffle=shuffle)
+        self.lengths = dataset.lengths
+        self.batch_size = batch_size
+        self.boundaries = boundaries
+
+        self.buckets, self.num_samples_per_bucket = self._create_buckets()
+        self.total_size = sum(self.num_samples_per_bucket)
+        self.num_samples = self.total_size // self.num_replicas
+
+    def _create_buckets(self):
+        buckets = [[] for _ in range(len(self.boundaries) - 1)]
+        for i in range(len(self.lengths)):
+            length = self.lengths[i]
+            idx_bucket = self._bisect(length)
+            if idx_bucket != -1:
+                buckets[idx_bucket].append(i)
+
+        try:
+            for i in range(len(buckets) - 1, 0, -1):
+                if len(buckets[i]) == 0:
+                    buckets.pop(i)
+                    self.boundaries.pop(i + 1)
+            assert all(len(bucket) > 0 for bucket in buckets)
+        # When one bucket is not traversed
+        except Exception as e:
+            print("Bucket warning ", e)
+            for i in range(len(buckets) - 1, -1, -1):
+                if len(buckets[i]) == 0:
+                    buckets.pop(i)
+                    self.boundaries.pop(i + 1)
+
+        num_samples_per_bucket = []
+        for i in range(len(buckets)):
+            len_bucket = len(buckets[i])
+            total_batch_size = self.num_replicas * self.batch_size
+            rem = (
+                total_batch_size - (len_bucket % total_batch_size)
+            ) % total_batch_size
+            num_samples_per_bucket.append(len_bucket + rem)
+        return buckets, num_samples_per_bucket
+
+    def __iter__(self):
+        # deterministically shuffle based on epoch
+        g = torch.Generator()
+        g.manual_seed(self.epoch)
+
+        indices = []
+        if self.shuffle:
+            for bucket in self.buckets:
+                indices.append(torch.randperm(len(bucket), generator=g).tolist())
+        else:
+            for bucket in self.buckets:
+                indices.append(list(range(len(bucket))))
+
+        batches = []
+        for i in range(len(self.buckets)):
+            bucket = self.buckets[i]
+            len_bucket = len(bucket)
+            if len_bucket == 0:
+                continue
+            ids_bucket = indices[i]
+            num_samples_bucket = self.num_samples_per_bucket[i]
+
+            # add extra samples to make it evenly divisible
+            rem = num_samples_bucket - len_bucket
+            ids_bucket = (
+                ids_bucket
+                + ids_bucket * (rem // len_bucket)
+                + ids_bucket[: (rem % len_bucket)]
+            )
+
+            # subsample
+            ids_bucket = ids_bucket[self.rank :: self.num_replicas]
+
+            # batching
+            for j in range(len(ids_bucket) // self.batch_size):
+                batch = [
+                    bucket[idx]
+                    for idx in ids_bucket[
+                        j * self.batch_size : (j + 1) * self.batch_size
+                    ]
+                ]
+                batches.append(batch)
+
+        if self.shuffle:
+            batch_ids = torch.randperm(len(batches), generator=g).tolist()
+            batches = [batches[i] for i in batch_ids]
+        self.batches = batches
+
+        assert len(self.batches) * self.batch_size == self.num_samples
+        return iter(self.batches)
+
+    def _bisect(self, x, lo=0, hi=None):
+        if hi is None:
+            hi = len(self.boundaries) - 1
+
+        if hi > lo:
+            mid = (hi + lo) // 2
+            if self.boundaries[mid] < x and x <= self.boundaries[mid + 1]:
+                return mid
+            elif x <= self.boundaries[mid]:
+                return self._bisect(x, lo, mid)
+            else:
+                return self._bisect(x, mid + 1, hi)
+        else:
+            return -1
+
+    def __len__(self):
+        return self.num_samples // self.batch_size

BanG-Dream-VITS/default_config.yml

@@ -0,0 +1,177 @@
+# 全局配置
+# 对于希望在同一时间使用多个配置文件的情况，例如两个GPU同时跑两个训练集：通过环境变量指定配置文件，不指定则默认为./config.yml
+
+# 拟提供通用路径配置，统一存放数据，避免数据放得很乱
+# 每个数据集与其对应的模型存放至统一路径下，后续所有的路径配置均为相对于datasetPath的路径
+# 不填或者填空则路径为相对于项目根目录的路径
+dataset_path: "Data/"
+
+# 模型镜像源，默认huggingface，使用openi镜像源需指定openi_token
+mirror: ""
+openi_token: ""  # openi token
+
+# resample 音频重采样配置
+# 注意， “:” 后需要加空格
+resample:
+  # 目标重采样率
+  sampling_rate: 44100
+  # 音频文件输入路径，重采样会将该路径下所有.wav音频文件重采样
+  # 请填入相对于datasetPath的相对路径
+  in_dir: "audios/raw" # 相对于根目录的路径为 /datasetPath/in_dir
+  # 音频文件重采样后输出路径
+  out_dir: "audios/wavs"
+
+
+# preprocess_text 数据集预处理相关配置
+# 注意， “:” 后需要加空格
+preprocess_text:
+  # 原始文本文件路径，文本格式应为{wav_path}|{speaker_name}|{language}|{text}。
+  transcription_path: "filelists/你的数据集文本.list"
+  # 数据清洗后文本路径，可以不填。不填则将在原始文本目录生成
+  cleaned_path: ""
+  # 训练集路径
+  train_path: "filelists/train.list"
+  # 验证集路径
+  val_path: "filelists/val.list"
+  # 配置文件路径
+  config_path: "config.json"
+  # 每个语言的验证集条数
+  val_per_lang: 4
+  # 验证集最大条数，多于的会被截断并放到训练集中
+  max_val_total: 12
+  # 是否进行数据清洗
+  clean: true
+
+
+# bert_gen 相关配置
+# 注意， “:” 后需要加空格
+bert_gen:
+  # 训练数据集配置文件路径
+  config_path: "config.json"
+  # 并行数
+  num_processes: 4
+  # 使用设备：可选项 "cuda" 显卡推理，"cpu" cpu推理
+  # 该选项同时决定了get_bert_feature的默认设备
+  device: "cuda"
+  # 使用多卡推理
+  use_multi_device: false
+
+# emo_gen 相关配置
+# 注意， “:” 后需要加空格
+emo_gen:
+  # 训练数据集配置文件路径
+  config_path: "config.json"
+  # 并行数
+  num_processes: 4
+  # 使用设备：可选项 "cuda" 显卡推理，"cpu" cpu推理
+  device: "cuda"
+  # 使用多卡推理
+  use_multi_device: false
+
+# train 训练配置
+# 注意， “:” 后需要加空格
+train_ms:
+  env:
+    MASTER_ADDR: "localhost"
+    MASTER_PORT: 10086
+    WORLD_SIZE: 1
+    LOCAL_RANK: 0
+    RANK: 0
+    # 可以填写任意名的环境变量
+    # THE_ENV_VAR_YOU_NEED_TO_USE: "1234567"
+  # 底模设置
+  base:
+    use_base_model: false
+    repo_id: "Stardust_minus/Bert-VITS2"
+    model_image: "Bert-VITS2_2.2-Clap底模" # openi网页的模型名
+  # 训练模型存储目录：与旧版本的区别，原先数据集是存放在logs/model_name下的，现在改为统一存放在Data/你的数据集/models下
+  model: "models"
+  # 配置文件路径
+  config_path: "config.json"
+  # 训练使用的worker，不建议超过CPU核心数
+  num_workers: 16
+  # 关闭此项可以节约接近50%的磁盘空间，但是可能导致实际训练速度变慢和更高的CPU使用率。
+  spec_cache: True
+  # 保存的检查点数量，多于此数目的权重会被删除来节省空间。
+  keep_ckpts: 8
+
+
+# webui webui配置
+# 注意， “:” 后需要加空格
+webui:
+  # 推理设备
+  device: "cuda"
+  # 模型路径
+  model: "models/G_8000.pth"
+  # 配置文件路径
+  config_path: "config.json"
+  # 端口号
+  port: 7860
+  # 是否公开部署，对外网开放
+  share: false
+  # 是否开启debug模式
+  debug: false
+  # 语种识别库，可选langid, fastlid
+  language_identification_library: "langid"
+
+
+# server-fastapi配置
+# 注意， “:” 后需要加空格
+# 注意，本配置下的所有配置均为相对于根目录的路径
+server:
+  # 端口号
+  port: 5000
+  # 模型默认使用设备：但是当前并没有实现这个配置。
+  device: "cuda"
+  # 需要加载的所有模型的配置，可以填多个模型，也可以不填模型，等网页成功后手动加载模型
+  # 不加载模型的配置格式：删除默认给的两个模型配置，给models赋值 [ ]，也就是空列表。参考模型2的speakers 即 models: [ ]
+  # 注意，所有模型都必须正确配置model与config的路径，空路径会导致加载错误。
+  # 也可以不填模型，等网页加载成功后手动填写models。
+  models:
+    - # 模型的路径
+      model: ""
+      # 模型config.json的路径
+      config: ""
+      # 模型使用设备，若填写则会覆盖默认配置
+      device: "cuda"
+      # 模型默认使用的语言
+      language: "ZH"
+      # 模型人物默认参数
+      # 不必填写所有人物，不填的使用默认值
+      # 暂时不用填写，当前尚未实现按人区分配置
+      speakers:
+        - speaker: "科比"
+          sdp_ratio: 0.2
+          noise_scale: 0.6
+          noise_scale_w: 0.8
+          length_scale: 1
+        - speaker: "五条悟"
+          sdp_ratio: 0.3
+          noise_scale: 0.7
+          noise_scale_w: 0.8
+          length_scale: 0.5
+        - speaker: "安倍晋三"
+          sdp_ratio: 0.2
+          noise_scale: 0.6
+          noise_scale_w: 0.8
+          length_scale: 1.2
+    - # 模型的路径
+      model: ""
+      # 模型config.json的路径
+      config: ""
+      # 模型使用设备，若填写则会覆盖默认配置
+      device: "cpu"
+      # 模型默认使用的语言
+      language: "JP"
+      # 模型人物默认参数
+      # 不必填写所有人物，不填的使用默认值
+      speakers: [ ] # 也可以不填
+
+# 百度翻译开放平台 api配置
+# api接入文档 https://api.fanyi.baidu.com/doc/21
+# 请不要在github等网站公开分享你的app id 与 key
+translate:
+  # 你的APPID
+  "app_key": ""
+  # 你的密钥
+  "secret_key": ""

BanG-Dream-VITS/emo_gen.py

@@ -0,0 +1,174 @@
+import argparse
+import os
+from pathlib import Path
+
+import librosa
+import numpy as np
+import torch
+import torch.nn as nn
+from torch.utils.data import Dataset
+from torch.utils.data import DataLoader, Dataset
+from tqdm import tqdm
+from transformers import Wav2Vec2Processor
+from transformers.models.wav2vec2.modeling_wav2vec2 import (
+    Wav2Vec2Model,
+    Wav2Vec2PreTrainedModel,
+)
+import sys
+import utils
+from config import config
+
+
+class RegressionHead(nn.Module):
+    r"""Classification head."""
+
+    def __init__(self, config):
+        super().__init__()
+
+        self.dense = nn.Linear(config.hidden_size, config.hidden_size)
+        self.dropout = nn.Dropout(config.final_dropout)
+        self.out_proj = nn.Linear(config.hidden_size, config.num_labels)
+
+    def forward(self, features, **kwargs):
+        x = features
+        x = self.dropout(x)
+        x = self.dense(x)
+        x = torch.tanh(x)
+        x = self.dropout(x)
+        x = self.out_proj(x)
+
+        return x
+
+
+class EmotionModel(Wav2Vec2PreTrainedModel):
+    r"""Speech emotion classifier."""
+
+    def __init__(self, config):
+        super().__init__(config)
+
+        self.config = config
+        self.wav2vec2 = Wav2Vec2Model(config)
+        self.classifier = RegressionHead(config)
+        self.init_weights()
+
+    def forward(
+        self,
+        input_values,
+    ):
+        outputs = self.wav2vec2(input_values)
+        hidden_states = outputs[0]
+        hidden_states = torch.mean(hidden_states, dim=1)
+        logits = self.classifier(hidden_states)
+
+        return hidden_states, logits
+
+
+class AudioDataset(Dataset):
+    def __init__(self, list_of_wav_files, sr, processor):
+        self.list_of_wav_files = list_of_wav_files
+        self.processor = processor
+        self.sr = sr
+
+    def __len__(self):
+        return len(self.list_of_wav_files)
+
+    def __getitem__(self, idx):
+        wav_file = self.list_of_wav_files[idx]
+        audio_data, _ = librosa.load(wav_file, sr=self.sr)
+        processed_data = self.processor(audio_data, sampling_rate=self.sr)[
+            "input_values"
+        ][0]
+        return torch.from_numpy(processed_data)
+
+
+def process_func(
+    x: np.ndarray,
+    sampling_rate: int,
+    model: EmotionModel,
+    processor: Wav2Vec2Processor,
+    device: str,
+    embeddings: bool = False,
+) -> np.ndarray:
+    device = (
+        "cuda:0"
+        if torch.cuda.is_available()
+        else (
+            "mps"
+            if sys.platform == "darwin" and torch.backends.mps.is_available()
+            else "cpu"
+        )
+    )
+    r"""Predict emotions or extract embeddings from raw audio signal."""
+    model = model.to(device)
+    y = processor(x, sampling_rate=sampling_rate)
+    y = y["input_values"][0]
+    y = torch.from_numpy(y).unsqueeze(0).to(device)
+
+    # run through model
+    with torch.no_grad():
+        y = model(y)[0 if embeddings else 1]
+
+    # convert to numpy
+    y = y.detach().cpu().numpy()
+
+    return y
+
+
+
+def get_emo(path):
+    wav, sr = librosa.load(path, 16000)
+    device = config.bert_gen_config.device
+    print("successfully generate the emo vec")
+    return process_func(
+        np.expand_dims(wav, 0).astype(np.float),
+        sr,
+        model,
+        processor,
+        device,
+        embeddings=True,
+    ).squeeze(0)
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser()
+    parser.add_argument(
+        "-c", "--config", type=str, default=config.bert_gen_config.config_path
+    )
+    parser.add_argument(
+        "--num_processes", type=int, default=config.bert_gen_config.num_processes
+    )
+    args, _ = parser.parse_known_args()
+    config_path = args.config
+    hps = utils.get_hparams_from_file(config_path)
+
+    device = config.bert_gen_config.device
+
+    model_name = "./emotional/wav2vec2-large-robust-12-ft-emotion-msp-dim"
+    REPO_ID = "audeering/wav2vec2-large-robust-12-ft-emotion-msp-dim"
+    if not Path(model_name).joinpath("pytorch_model.bin").exists():
+        utils.download_emo_models(config.mirror, REPO_ID, model_name)
+
+    processor = Wav2Vec2Processor.from_pretrained(model_name)
+    model = EmotionModel.from_pretrained(model_name).to(device)
+
+    lines = []
+    with open(hps.data.training_files, encoding="utf-8") as f:
+        lines.extend(f.readlines())
+
+    with open(hps.data.validation_files, encoding="utf-8") as f:
+        lines.extend(f.readlines())
+
+    wavnames = [line.split("|")[0] for line in lines]
+    dataset = AudioDataset(wavnames, 16000, processor)
+    data_loader = DataLoader(dataset, batch_size=1, shuffle=False, num_workers=16)
+
+    with torch.no_grad():
+        for i, data in tqdm(enumerate(data_loader), total=len(data_loader)):
+            wavname = wavnames[i]
+            emo_path = wavname.replace(".wav", ".emo.npy")
+            if os.path.exists(emo_path):
+                continue
+            emb = model(data.to(device))[0].detach().cpu().numpy()
+            np.save(emo_path, emb)
+
+    print("Emo vec 生成完毕!")

BanG-Dream-VITS/empty_emo.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:07063411ab7d6e7aacfc73c582616c3fbc8fdf518b20d42d8be77bc9caf6fab9
+size 3238

BanG-Dream-VITS/export_onnx.py

@@ -0,0 +1,14 @@
+from onnx_modules import export_onnx
+import os
+
+if __name__ == "__main__":
+    export_path = "BertVits2.2PT"
+    model_path = "model\\G_0.pth"
+    config_path = "model\\config.json"
+    novq = False
+    dev = False
+    if not os.path.exists("onnx"):
+        os.makedirs("onnx")
+    if not os.path.exists(f"onnx/{export_path}"):
+        os.makedirs(f"onnx/{export_path}")
+    export_onnx(export_path, model_path, config_path, novq, dev)

BanG-Dream-VITS/filelists/esd.list

@@ -0,0 +1,3 @@
+Example:
+{wav_path}|{speaker_name}|{language}|{text}
+派蒙_1.wav|派蒙|ZH|前面的区域，以后再来探索吧！

BanG-Dream-VITS/filelists/sample.list

@@ -0,0 +1,3 @@
+Example:
+{wav_path}|{speaker_name}|{language}|{text}
+派蒙_1.wav|派蒙|ZH|前面的区域，以后再来探索吧！

BanG-Dream-VITS/filelists/val.list

@@ -0,0 +1,8 @@
+/content/drive/MyDrive/Mygo/event235-15-063.wav|燈|JP|うん……愛音ちゃん……ありがとう|_ u N . . . . . . a i o t o ch a N . . . . . . a r i g a t o _|0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0|1 2 1 1 1 1 1 1 2 3 3 1 1 1 1 1 1 7 1
+/content/drive/MyDrive/Mygo/event235-29-022.wav|燈|JP|私……祥ちゃんも……みんな,傷つけて……|_ w a t a k u sh i . . . . . . s a ch i ch a N m o . . . . . . m i N n a , k i z u ts u k e t e . . . . . . _|0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0|1 8 1 1 1 1 1 1 4 3 2 1 1 1 1 1 1 5 1 8 2 1 1 1 1 1 1 1
+/content/drive/MyDrive/Mygo/event235-30-058.wav|燈|JP|無理じゃない!|_ m u r i j a n a i ! _|0 0 0 0 0 0 0 0 0 0 0 0|1 4 2 3 1 1
+/content/drive/MyDrive/Mygo/event235-32-011.wav|燈|JP|僕が手を離してしまった|_ b o k u g a t e o h a n a sh i t e sh i m a q t a _|0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0|1 4 2 2 1 3 3 2 5 2 1
+/content/drive/MyDrive/Mygo/event235-05-001.wav|そよ|JP|ごきげんよう,睦ちゃん.今日も菜園のお世話してるの?|_ g o k i g e N y o , b o k u ch a N . ky o m o s a i e N n o o s e w a sh i t e r u n o ? _|0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0|1 2 3 2 2 1 4 3 1 2 2 3 2 2 1 4 2 4 2 1 1
+/content/drive/MyDrive/Mygo/event235-28-016.wav|そよ|JP|心のどこかで,ここは自分がいる場所じゃないような気がして|_ k o k o r o n o d o k o k a d e , k o k o w a j i b u N g a i r u b a sh o j a n a i y o n a k i g a sh i t e _|0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0|1 6 2 4 2 2 1 4 2 5 2 3 4 2 3 2 2 2 2 2 2 1
+/content/drive/MyDrive/Mygo/event235-39-005.wav|そよ|JP|やりたいって言ったの誰だっけ.ライブ,明日だけど|_ y a r i t a i q t e i q t a n o d a r e d a q k e . r a i b u , a s u d a k e d o _|0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0|1 4 3 3 2 2 2 4 2 2 1 1 5 1 3 2 4 1
+/content/drive/MyDrive/Mygo/event235-27-053.wav|そよ|JP|……でも,私,本気で|_ . . . . . . d e m o , w a t a k u sh i , h o N k i d e _|0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0|1 1 1 1 1 1 1 2 2 1 8 1 5 2 1

BanG-Dream-VITS/hiyoriUI.py

@@ -0,0 +1,725 @@
+"""
+api服务，网页后端 多版本多模型 fastapi实现
+原 server_fastapi
+"""
+import logging
+import gc
+import random
+import librosa
+import gradio
+import numpy as np
+import utils
+from fastapi import FastAPI, Query, Request, File, UploadFile, Form
+from fastapi.responses import Response, FileResponse
+from fastapi.staticfiles import StaticFiles
+from io import BytesIO
+from scipy.io import wavfile
+import uvicorn
+import torch
+import webbrowser
+import psutil
+import GPUtil
+from typing import Dict, Optional, List, Set, Union, Tuple
+import os
+from tools.log import logger
+from urllib.parse import unquote
+
+from infer import infer, get_net_g, latest_version
+import tools.translate as trans
+from tools.sentence import split_by_language
+from re_matching import cut_sent
+
+
+from config import config
+
+os.environ["TOKENIZERS_PARALLELISM"] = "false"
+
+
+class Model:
+    """模型封装类"""
+
+    def __init__(self, config_path: str, model_path: str, device: str, language: str):
+        self.config_path: str = os.path.normpath(config_path)
+        self.model_path: str = os.path.normpath(model_path)
+        self.device: str = device
+        self.language: str = language
+        self.hps = utils.get_hparams_from_file(config_path)
+        self.spk2id: Dict[str, int] = self.hps.data.spk2id  # spk - id 映射字典
+        self.id2spk: Dict[int, str] = dict()  # id - spk 映射字典
+        for speaker, speaker_id in self.hps.data.spk2id.items():
+            self.id2spk[speaker_id] = speaker
+        self.version: str = (
+            self.hps.version if hasattr(self.hps, "version") else latest_version
+        )
+        self.net_g = get_net_g(
+            model_path=model_path,
+            version=self.version,
+            device=device,
+            hps=self.hps,
+        )
+
+    def to_dict(self) -> Dict[str, any]:
+        return {
+            "config_path": self.config_path,
+            "model_path": self.model_path,
+            "device": self.device,
+            "language": self.language,
+            "spk2id": self.spk2id,
+            "id2spk": self.id2spk,
+            "version": self.version,
+        }
+
+
+class Models:
+    def __init__(self):
+        self.models: Dict[int, Model] = dict()
+        self.num = 0
+        # spkInfo[角色名][模型id] = 角色id
+        self.spk_info: Dict[str, Dict[int, int]] = dict()
+        self.path2ids: Dict[str, Set[int]] = dict()  # 路径指向的model的id
+
+    def init_model(
+        self, config_path: str, model_path: str, device: str, language: str
+    ) -> int:
+        """
+        初始化并添加一个模型
+
+        :param config_path: 模型config.json路径
+        :param model_path: 模型路径
+        :param device: 模型推理使用设备
+        :param language: 模型推理默认语言
+        """
+        # 若文件不存在则不进行加载
+        if not os.path.isfile(model_path):
+            if model_path != "":
+                logger.warning(f"模型文件{model_path} 不存在，不进行初始化")
+            return self.num
+        if not os.path.isfile(config_path):
+            if config_path != "":
+                logger.warning(f"配置文件{config_path} 不存在，不进行初始化")
+            return self.num
+
+        # 若路径中的模型已存在，则不添加模型，若不存在，则进行初始化。
+        model_path = os.path.realpath(model_path)
+        if model_path not in self.path2ids.keys():
+            self.path2ids[model_path] = {self.num}
+            self.models[self.num] = Model(
+                config_path=config_path,
+                model_path=model_path,
+                device=device,
+                language=language,
+            )
+            logger.success(f"添加模型{model_path}，使用配置文件{os.path.realpath(config_path)}")
+        else:
+            # 获取一个指向id
+            m_id = next(iter(self.path2ids[model_path]))
+            self.models[self.num] = self.models[m_id]
+            self.path2ids[model_path].add(self.num)
+            logger.success("模型已存在，添加模型引用。")
+        # 添加角色信息
+        for speaker, speaker_id in self.models[self.num].spk2id.items():
+            if speaker not in self.spk_info.keys():
+                self.spk_info[speaker] = {self.num: speaker_id}
+            else:
+                self.spk_info[speaker][self.num] = speaker_id
+        # 修改计数
+        self.num += 1
+        return self.num - 1
+
+    def del_model(self, index: int) -> Optional[int]:
+        """删除对应序号的模型，若不存在则返回None"""
+        if index not in self.models.keys():
+            return None
+        # 删除角色信息
+        for speaker, speaker_id in self.models[index].spk2id.items():
+            self.spk_info[speaker].pop(index)
+            if len(self.spk_info[speaker]) == 0:
+                # 若对应角色的所有模型都被删除，则清除该角色信息
+                self.spk_info.pop(speaker)
+        # 删除路径信息
+        model_path = os.path.realpath(self.models[index].model_path)
+        self.path2ids[model_path].remove(index)
+        if len(self.path2ids[model_path]) == 0:
+            self.path2ids.pop(model_path)
+            logger.success(f"删除模型{model_path}, id = {index}")
+        else:
+            logger.success(f"删除模型引用{model_path}, id = {index}")
+        # 删除模型
+        self.models.pop(index)
+        gc.collect()
+        if torch.cuda.is_available():
+            torch.cuda.empty_cache()
+        return index
+
+    def get_models(self):
+        """获取所有模型"""
+        return self.models
+
+
+if __name__ == "__main__":
+    app = FastAPI()
+    app.logger = logger
+    # 挂载静态文件
+    logger.info("开始挂载网页页面")
+    StaticDir: str = "./Web"
+    if not os.path.isdir(StaticDir):
+        logger.warning(
+            "缺少网页资源，无法开启网页页面，如有需要请在 https://github.com/jiangyuxiaoxiao/Bert-VITS2-UI 或者Bert-VITS对应版本的release页面下载"
+        )
+    else:
+        dirs = [fir.name for fir in os.scandir(StaticDir) if fir.is_dir()]
+        files = [fir.name for fir in os.scandir(StaticDir) if fir.is_dir()]
+        for dirName in dirs:
+            app.mount(
+                f"/{dirName}",
+                StaticFiles(directory=f"./{StaticDir}/{dirName}"),
+                name=dirName,
+            )
+    loaded_models = Models()
+    # 加载模型
+    logger.info("开始加载模型")
+    models_info = config.server_config.models
+    for model_info in models_info:
+        loaded_models.init_model(
+            config_path=model_info["config"],
+            model_path=model_info["model"],
+            device=model_info["device"],
+            language=model_info["language"],
+        )
+
+    @app.get("/")
+    async def index():
+        return FileResponse("./Web/index.html")
+
+    async def _voice(
+        text: str,
+        model_id: int,
+        speaker_name: str,
+        speaker_id: int,
+        sdp_ratio: float,
+        noise: float,
+        noisew: float,
+        length: float,
+        language: str,
+        auto_translate: bool,
+        auto_split: bool,
+        emotion: Optional[Union[int, str]] = None,
+        reference_audio=None,
+        style_text: Optional[str] = None,
+        style_weight: float = 0.7,
+    ) -> Union[Response, Dict[str, any]]:
+        """TTS实现函数"""
+
+        # 检查
+        # 检查模型是否存在
+        if model_id not in loaded_models.models.keys():
+            logger.error(f"/voice 请求错误：模型model_id={model_id}未加载")
+            return {"status": 10, "detail": f"模型model_id={model_id}未加载"}
+        # 检查是否提供speaker
+        if speaker_name is None and speaker_id is None:
+            logger.error("/voice 请求错误：推理请求未提供speaker_name或speaker_id")
+            return {"status": 11, "detail": "请提供speaker_name或speaker_id"}
+        elif speaker_name is None:
+            # 检查speaker_id是否存在
+            if speaker_id not in loaded_models.models[model_id].id2spk.keys():
+                logger.error(f"/voice 请求错误：角色speaker_id={speaker_id}不存在")
+                return {"status": 12, "detail": f"角色speaker_id={speaker_id}不存在"}
+            speaker_name = loaded_models.models[model_id].id2spk[speaker_id]
+        # 检查speaker_name是否存在
+        if speaker_name not in loaded_models.models[model_id].spk2id.keys():
+            logger.error(f"/voice 请求错误：角色speaker_name={speaker_name}不存在")
+            return {"status": 13, "detail": f"角色speaker_name={speaker_name}不存在"}
+        # 未传入则使用默认语言
+        if language is None:
+            language = loaded_models.models[model_id].language
+        # 翻译会破坏mix结构，auto也会变得无意义。不要在这两个模式下使用
+        if auto_translate:
+            if language == "auto" or language == "mix":
+                logger.error(
+                    f"/voice 请求错误：请勿同时使用language = {language}与auto_translate模式"
+                )
+                return {
+                    "status": 20,
+                    "detail": f"请勿同时使用language = {language}与auto_translate模式",
+                }
+            text = trans.translate(Sentence=text, to_Language=language.lower())
+        if reference_audio is not None:
+            ref_audio = BytesIO(await reference_audio.read())
+            # 2.2 适配
+            if loaded_models.models[model_id].version == "2.2":
+                ref_audio, _ = librosa.load(ref_audio, 48000)
+        else:
+            ref_audio = reference_audio
+
+        # 改动：增加使用 || 对文本进行主动切分
+        # 切分优先级： || → auto/mix → auto_split
+        text2 = text.replace("\n", "").lstrip()
+        texts: List[str] = text2.split("||")
+
+        # 对于mix和auto的说明：出于版本兼容性���考虑，暂时无法使用multilang的方式进行推理
+        if language == "MIX":
+            text_language_speakers: List[Tuple[str, str, str]] = []
+            for _text in texts:
+                speaker_pieces = _text.split("[")  # 按说话人分割多块
+                for speaker_piece in speaker_pieces:
+                    if speaker_piece == "":
+                        continue
+                    speaker_piece2 = speaker_piece.split("]")
+                    if len(speaker_piece2) != 2:
+                        return {
+                            "status": 21,
+                            "detail": "MIX语法错误",
+                        }
+                    speaker = speaker_piece2[0].strip()
+                    lang_pieces = speaker_piece2[1].split("<")
+                    for lang_piece in lang_pieces:
+                        if lang_piece == "":
+                            continue
+                        lang_piece2 = lang_piece.split(">")
+                        if len(lang_piece2) != 2:
+                            return {
+                                "status": 21,
+                                "detail": "MIX语法错误",
+                            }
+                        lang = lang_piece2[0].strip()
+                        if lang.upper() not in ["ZH", "EN", "JP"]:
+                            return {
+                                "status": 21,
+                                "detail": "MIX语法错误",
+                            }
+                        t = lang_piece2[1]
+                        text_language_speakers.append((t, lang.upper(), speaker))
+
+        elif language == "AUTO":
+            text_language_speakers: List[Tuple[str, str, str]] = [
+                (final_text, language.upper().replace("JA", "JP"), speaker_name)
+                for sub_list in [
+                    split_by_language(_text, target_languages=["zh", "ja", "en"])
+                    for _text in texts
+                    if _text != ""
+                ]
+                for final_text, language in sub_list
+                if final_text != ""
+            ]
+        else:
+            text_language_speakers: List[Tuple[str, str, str]] = [
+                (_text, language, speaker_name) for _text in texts if _text != ""
+            ]
+
+        if auto_split:
+            text_language_speakers: List[Tuple[str, str, str]] = [
+                (final_text, lang, speaker)
+                for _text, lang, speaker in text_language_speakers
+                for final_text in cut_sent(_text)
+            ]
+
+        audios = []
+        with torch.no_grad():
+            for _text, lang, speaker in text_language_speakers:
+                audios.append(
+                    infer(
+                        text=_text,
+                        sdp_ratio=sdp_ratio,
+                        noise_scale=noise,
+                        noise_scale_w=noisew,
+                        length_scale=length,
+                        sid=speaker,
+                        language=lang,
+                        hps=loaded_models.models[model_id].hps,
+                        net_g=loaded_models.models[model_id].net_g,
+                        device=loaded_models.models[model_id].device,
+                        emotion=emotion,
+                        reference_audio=ref_audio,
+                        style_text=style_text,
+                        style_weight=style_weight,
+                    )
+                )
+                # audios.append(np.zeros(int(44100 * 0.2)))
+            # audios.pop()
+            audio = np.concatenate(audios)
+            audio = gradio.processing_utils.convert_to_16_bit_wav(audio)
+        with BytesIO() as wavContent:
+            wavfile.write(
+                wavContent, loaded_models.models[model_id].hps.data.sampling_rate, audio
+            )
+            response = Response(content=wavContent.getvalue(), media_type="audio/wav")
+            return response
+
+    @app.post("/voice")
+    async def voice(
+        request: Request,  # fastapi自动注入
+        text: str = Form(...),
+        model_id: int = Query(..., description="模型ID"),  # 模型序号
+        speaker_name: str = Query(
+            None, description="说话人名"
+        ),  # speaker_name与 speaker_id二者选其一
+        speaker_id: int = Query(None, description="说话人id，与speaker_name二选一"),
+        sdp_ratio: float = Query(0.2, description="SDP/DP混合比"),
+        noise: float = Query(0.2, description="感情"),
+        noisew: float = Query(0.9, description="音素长度"),
+        length: float = Query(1, description="语速"),
+        language: str = Query(None, description="语言"),  # 若不指定使用语言则使用默认值
+        auto_translate: bool = Query(False, description="自动翻译"),
+        auto_split: bool = Query(False, description="自动切分"),
+        emotion: Optional[Union[int, str]] = Query(None, description="emo"),
+        reference_audio: UploadFile = File(None),
+        style_text: Optional[str] = Form(None, description="风格文本"),
+        style_weight: float = Query(0.7, description="风格权重"),
+    ):
+        """语音接口，若需要上传参考音频请仅使用post请求"""
+        logger.info(
+            f"{request.client.host}:{request.client.port}/voice  { unquote(str(request.query_params) )} text={text}"
+        )
+        return await _voice(
+            text=text,
+            model_id=model_id,
+            speaker_name=speaker_name,
+            speaker_id=speaker_id,
+            sdp_ratio=sdp_ratio,
+            noise=noise,
+            noisew=noisew,
+            length=length,
+            language=language,
+            auto_translate=auto_translate,
+            auto_split=auto_split,
+            emotion=emotion,
+            reference_audio=reference_audio,
+            style_text=style_text,
+            style_weight=style_weight,
+        )
+
+    @app.get("/voice")
+    async def voice(
+        request: Request,  # fastapi自动注入
+        text: str = Query(..., description="输入文字"),
+        model_id: int = Query(..., description="模型ID"),  # 模型序号
+        speaker_name: str = Query(
+            None, description="说话人名"
+        ),  # speaker_name与 speaker_id二者选其一
+        speaker_id: int = Query(None, description="说话人id，与speaker_name二选一"),
+        sdp_ratio: float = Query(0.2, description="SDP/DP混合比"),
+        noise: float = Query(0.2, description="感情"),
+        noisew: float = Query(0.9, description="音素长度"),
+        length: float = Query(1, description="语速"),
+        language: str = Query(None, description="语言"),  # 若不指定使用语言则使用默认值
+        auto_translate: bool = Query(False, description="自动翻译"),
+        auto_split: bool = Query(False, description="自动切分"),
+        emotion: Optional[Union[int, str]] = Query(None, description="emo"),
+        style_text: Optional[str] = Query(None, description="风格文本"),
+        style_weight: float = Query(0.7, description="风格权重"),
+    ):
+        """语音接口，不建议使用"""
+        logger.info(
+            f"{request.client.host}:{request.client.port}/voice  { unquote(str(request.query_params) )}"
+        )
+        return await _voice(
+            text=text,
+            model_id=model_id,
+            speaker_name=speaker_name,
+            speaker_id=speaker_id,
+            sdp_ratio=sdp_ratio,
+            noise=noise,
+            noisew=noisew,
+            length=length,
+            language=language,
+            auto_translate=auto_translate,
+            auto_split=auto_split,
+            emotion=emotion,
+            style_text=style_text,
+            style_weight=style_weight,
+        )
+
+    @app.get("/models/info")
+    def get_loaded_models_info(request: Request):
+        """获取已加载模型信息"""
+
+        result: Dict[str, Dict] = dict()
+        for key, model in loaded_models.models.items():
+            result[str(key)] = model.to_dict()
+        return result
+
+    @app.get("/models/delete")
+    def delete_model(
+        request: Request, model_id: int = Query(..., description="删除模型id")
+    ):
+        """删除指定模型"""
+        logger.info(
+            f"{request.client.host}:{request.client.port}/models/delete  { unquote(str(request.query_params) )}"
+        )
+        result = loaded_models.del_model(model_id)
+        if result is None:
+            logger.error(f"/models/delete 模型删除错误：模型{model_id}不存在，删除失败")
+            return {"status": 14, "detail": f"模型{model_id}不存在，删除失败"}
+
+        return {"status": 0, "detail": "删除成功"}
+
+    @app.get("/models/add")
+    def add_model(
+        request: Request,
+        model_path: str = Query(..., description="添加模型路径"),
+        config_path: str = Query(
+            None, description="添加模型配置文件路径，不填则使用./config.json或../config.json"
+        ),
+        device: str = Query("cuda", description="推理使用设备"),
+        language: str = Query("ZH", description="模型默认语言"),
+    ):
+        """添加指定模型：允许重复添加相同路径模型，且不重复占用内存"""
+        logger.info(
+            f"{request.client.host}:{request.client.port}/models/add  { unquote(str(request.query_params) )}"
+        )
+        if config_path is None:
+            model_dir = os.path.dirname(model_path)
+            if os.path.isfile(os.path.join(model_dir, "config.json")):
+                config_path = os.path.join(model_dir, "config.json")
+            elif os.path.isfile(os.path.join(model_dir, "../config.json")):
+                config_path = os.path.join(model_dir, "../config.json")
+            else:
+                logger.error("/models/add 模型添加失败：未在模型所在目录以及上级目录找到config.json文件")
+                return {
+                    "status": 15,
+                    "detail": "查询未传���配置文件路径，同时默认路径./与../中不存在配置文件config.json。",
+                }
+        try:
+            model_id = loaded_models.init_model(
+                config_path=config_path,
+                model_path=model_path,
+                device=device,
+                language=language,
+            )
+        except Exception:
+            logging.exception("模型加载出错")
+            return {
+                "status": 16,
+                "detail": "模型加载出错，详细查看日志",
+            }
+        return {
+            "status": 0,
+            "detail": "模型添加成功",
+            "Data": {
+                "model_id": model_id,
+                "model_info": loaded_models.models[model_id].to_dict(),
+            },
+        }
+
+    def _get_all_models(root_dir: str = "Data", only_unloaded: bool = False):
+        """从root_dir搜索获取所有可用模型"""
+        result: Dict[str, List[str]] = dict()
+        files = os.listdir(root_dir) + ["."]
+        for file in files:
+            if os.path.isdir(os.path.join(root_dir, file)):
+                sub_dir = os.path.join(root_dir, file)
+                # 搜索 "sub_dir" 、 "sub_dir/models" 两个路径
+                result[file] = list()
+                sub_files = os.listdir(sub_dir)
+                model_files = []
+                for sub_file in sub_files:
+                    relpath = os.path.realpath(os.path.join(sub_dir, sub_file))
+                    if only_unloaded and relpath in loaded_models.path2ids.keys():
+                        continue
+                    if sub_file.endswith(".pth") and sub_file.startswith("G_"):
+                        if os.path.isfile(relpath):
+                            model_files.append(sub_file)
+                # 对模型文件按步数排序
+                model_files = sorted(
+                    model_files,
+                    key=lambda pth: int(pth.lstrip("G_").rstrip(".pth"))
+                    if pth.lstrip("G_").rstrip(".pth").isdigit()
+                    else 10**10,
+                )
+                result[file] = model_files
+                models_dir = os.path.join(sub_dir, "models")
+                model_files = []
+                if os.path.isdir(models_dir):
+                    sub_files = os.listdir(models_dir)
+                    for sub_file in sub_files:
+                        relpath = os.path.realpath(os.path.join(models_dir, sub_file))
+                        if only_unloaded and relpath in loaded_models.path2ids.keys():
+                            continue
+                        if sub_file.endswith(".pth") and sub_file.startswith("G_"):
+                            if os.path.isfile(os.path.join(models_dir, sub_file)):
+                                model_files.append(f"models/{sub_file}")
+                    # 对模型文件按步数排序
+                    model_files = sorted(
+                        model_files,
+                        key=lambda pth: int(pth.lstrip("models/G_").rstrip(".pth"))
+                        if pth.lstrip("models/G_").rstrip(".pth").isdigit()
+                        else 10**10,
+                    )
+                    result[file] += model_files
+                if len(result[file]) == 0:
+                    result.pop(file)
+
+        return result
+
+    @app.get("/models/get_unloaded")
+    def get_unloaded_models_info(
+        request: Request, root_dir: str = Query("Data", description="搜索根目录")
+    ):
+        """获取未加载模型"""
+        logger.info(
+            f"{request.client.host}:{request.client.port}/models/get_unloaded  { unquote(str(request.query_params) )}"
+        )
+        return _get_all_models(root_dir, only_unloaded=True)
+
+    @app.get("/models/get_local")
+    def get_local_models_info(
+        request: Request, root_dir: str = Query("Data", description="搜索根目录")
+    ):
+        """获取全部本地模型"""
+        logger.info(
+            f"{request.client.host}:{request.client.port}/models/get_local  { unquote(str(request.query_params) )}"
+        )
+        return _get_all_models(root_dir, only_unloaded=False)
+
+    @app.get("/status")
+    def get_status():
+        """获取电脑运行状态"""
+        cpu_percent = psutil.cpu_percent(interval=1)
+        memory_info = psutil.virtual_memory()
+        memory_total = memory_info.total
+        memory_available = memory_info.available
+        memory_used = memory_info.used
+        memory_percent = memory_info.percent
+        gpuInfo = []
+        devices = ["cpu"]
+        for i in range(torch.cuda.device_count()):
+            devices.append(f"cuda:{i}")
+        gpus = GPUtil.getGPUs()
+        for gpu in gpus:
+            gpuInfo.append(
+                {
+                    "gpu_id": gpu.id,
+                    "gpu_load": gpu.load,
+                    "gpu_memory": {
+                        "total": gpu.memoryTotal,
+                        "used": gpu.memoryUsed,
+                        "free": gpu.memoryFree,
+                    },
+                }
+            )
+        return {
+            "devices": devices,
+            "cpu_percent": cpu_percent,
+            "memory_total": memory_total,
+            "memory_available": memory_available,
+            "memory_used": memory_used,
+            "memory_percent": memory_percent,
+            "gpu": gpuInfo,
+        }
+
+    @app.get("/tools/translate")
+    def translate(
+        request: Request,
+        texts: str = Query(..., description="待翻译文本"),
+        to_language: str = Query(..., description="翻译目标语言"),
+    ):
+        """翻译"""
+        logger.info(
+            f"{request.client.host}:{request.client.port}/tools/translate  { unquote(str(request.query_params) )}"
+        )
+        return {"texts": trans.translate(Sentence=texts, to_Language=to_language)}
+
+    all_examples: Dict[str, Dict[str, List]] = dict()  # 存放示例
+
+    @app.get("/tools/random_example")
+    def random_example(
+        request: Request,
+        language: str = Query(None, description="指定语言，未指定则随机返回"),
+        root_dir: str = Query("Data", description="搜索根目录"),
+    ):
+        """
+        获取一个随机音频+文本，用于对比，音频会从本地目录随机选择。
+        """
+        logger.info(
+            f"{request.client.host}:{request.client.port}/tools/random_example  { unquote(str(request.query_params) )}"
+        )
+        global all_examples
+        # 数据初始化
+        if root_dir not in all_examples.keys():
+            all_examples[root_dir] = {"ZH": [], "JP": [], "EN": []}
+
+            examples = all_examples[root_dir]
+
+            # 从项目Data目录中搜索train/val.list
+            for root, directories, _files in os.walk(root_dir):
+                for file in _files:
+                    if file in ["train.list", "val.list"]:
+                        with open(
+                            os.path.join(root, file), mode="r", encoding="utf-8"
+                        ) as f:
+                            lines = f.readlines()
+                            for line in lines:
+                                data = line.split("|")
+                                if len(data) != 7:
+                                    continue
+                                # 音频存在 且语言为ZH/EN/JP
+                                if os.path.isfile(data[0]) and data[2] in [
+                                    "ZH",
+                                    "JP",
+                                    "EN",
+                                ]:
+                                    examples[data[2]].append(
+                                        {
+                                            "text": data[3],
+                                            "audio": data[0],
+                                            "speaker": data[1],
+                                        }
+                                    )
+
+        examples = all_examples[root_dir]
+        if language is None:
+            if len(examples["ZH"]) + len(examples["JP"]) + len(examples["EN"]) == 0:
+                return {"status": 17, "detail": "没有加载任何示例数据"}
+            else:
+                # 随机选一个
+                rand_num = random.randint(
+                    0,
+                    len(examples["ZH"]) + len(examples["JP"]) + len(examples["EN"]) - 1,
+                )
+                # ZH
+                if rand_num < len(examples["ZH"]):
+                    return {"status": 0, "Data": examples["ZH"][rand_num]}
+                # JP
+                if rand_num < len(examples["ZH"]) + len(examples["JP"]):
+                    return {
+                        "status": 0,
+                        "Data": examples["JP"][rand_num - len(examples["ZH"])],
+                    }
+                # EN
+                return {
+                    "status": 0,
+                    "Data": examples["EN"][
+                        rand_num - len(examples["ZH"]) - len(examples["JP"])
+                    ],
+                }
+
+        else:
+            if len(examples[language]) == 0:
+                return {"status": 17, "detail": f"没有加载任何{language}数据"}
+            return {
+                "status": 0,
+                "Data": examples[language][
+                    random.randint(0, len(examples[language]) - 1)
+                ],
+            }
+
+    @app.get("/tools/get_audio")
+    def get_audio(request: Request, path: str = Query(..., description="本地音频路径")):
+        logger.info(
+            f"{request.client.host}:{request.client.port}/tools/get_audio  { unquote(str(request.query_params) )}"
+        )
+        if not os.path.isfile(path):
+            logger.error(f"/tools/get_audio 获取音频错误：指定音频{path}不存在")
+            return {"status": 18, "detail": "指定音频不存在"}
+        if not path.lower().endswith(".wav"):
+            logger.error(f"/tools/get_audio 获取音频错误：音频{path}非wav文件")
+            return {"status": 19, "detail": "非wav格式文件"}
+        return FileResponse(path=path)
+
+    logger.warning("本地服务，请勿将服务端口暴露于外网")
+    logger.info(f"api文档地址 http://127.0.0.1:{config.server_config.port}/docs")
+    if os.path.isdir(StaticDir):
+        webbrowser.open(f"http://127.0.0.1:{config.server_config.port}")
+    uvicorn.run(
+        app, port=config.server_config.port, host="0.0.0.0", log_level="warning"
+    )