apptor commited on
Commit
b5c6ae8
·
verified ·
1 Parent(s): b18bc37

apptor-tiny-tts German v1 (step 82044, Vocos 60k)

Browse files
.gitattributes CHANGED
@@ -33,3 +33,5 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
33
  *.zip filter=lfs diff=lfs merge=lfs -text
34
  *.zst filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
 
 
 
33
  *.zip filter=lfs diff=lfs merge=lfs -text
34
  *.zst filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
36
+ samples/demo-without-vocos.wav filter=lfs diff=lfs merge=lfs -text
37
+ samples/demo.wav filter=lfs diff=lfs merge=lfs -text
LICENSE ADDED
@@ -0,0 +1,202 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+
2
+ Apache License
3
+ Version 2.0, January 2004
4
+ http://www.apache.org/licenses/
5
+
6
+ TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
7
+
8
+ 1. Definitions.
9
+
10
+ "License" shall mean the terms and conditions for use, reproduction,
11
+ and distribution as defined by Sections 1 through 9 of this document.
12
+
13
+ "Licensor" shall mean the copyright owner or entity authorized by
14
+ the copyright owner that is granting the License.
15
+
16
+ "Legal Entity" shall mean the union of the acting entity and all
17
+ other entities that control, are controlled by, or are under common
18
+ control with that entity. For the purposes of this definition,
19
+ "control" means (i) the power, direct or indirect, to cause the
20
+ direction or management of such entity, whether by contract or
21
+ otherwise, or (ii) ownership of fifty percent (50%) or more of the
22
+ outstanding shares, or (iii) beneficial ownership of such entity.
23
+
24
+ "You" (or "Your") shall mean an individual or Legal Entity
25
+ exercising permissions granted by this License.
26
+
27
+ "Source" form shall mean the preferred form for making modifications,
28
+ including but not limited to software source code, documentation
29
+ source, and configuration files.
30
+
31
+ "Object" form shall mean any form resulting from mechanical
32
+ transformation or translation of a Source form, including but
33
+ not limited to compiled object code, generated documentation,
34
+ and conversions to other media types.
35
+
36
+ "Work" shall mean the work of authorship, whether in Source or
37
+ Object form, made available under the License, as indicated by a
38
+ copyright notice that is included in or attached to the work
39
+ (an example is provided in the Appendix below).
40
+
41
+ "Derivative Works" shall mean any work, whether in Source or Object
42
+ form, that is based on (or derived from) the Work and for which the
43
+ editorial revisions, annotations, elaborations, or other modifications
44
+ represent, as a whole, an original work of authorship. For the purposes
45
+ of this License, Derivative Works shall not include works that remain
46
+ separable from, or merely link (or bind by name) to the interfaces of,
47
+ the Work and Derivative Works thereof.
48
+
49
+ "Contribution" shall mean any work of authorship, including
50
+ the original version of the Work and any modifications or additions
51
+ to that Work or Derivative Works thereof, that is intentionally
52
+ submitted to Licensor for inclusion in the Work by the copyright owner
53
+ or by an individual or Legal Entity authorized to submit on behalf of
54
+ the copyright owner. For the purposes of this definition, "submitted"
55
+ means any form of electronic, verbal, or written communication sent
56
+ to the Licensor or its representatives, including but not limited to
57
+ communication on electronic mailing lists, source code control systems,
58
+ and issue tracking systems that are managed by, or on behalf of, the
59
+ Licensor for the purpose of discussing and improving the Work, but
60
+ excluding communication that is conspicuously marked or otherwise
61
+ designated in writing by the copyright owner as "Not a Contribution."
62
+
63
+ "Contributor" shall mean Licensor and any individual or Legal Entity
64
+ on behalf of whom a Contribution has been received by Licensor and
65
+ subsequently incorporated within the Work.
66
+
67
+ 2. Grant of Copyright License. Subject to the terms and conditions of
68
+ this License, each Contributor hereby grants to You a perpetual,
69
+ worldwide, non-exclusive, no-charge, royalty-free, irrevocable
70
+ copyright license to reproduce, prepare Derivative Works of,
71
+ publicly display, publicly perform, sublicense, and distribute the
72
+ Work and such Derivative Works in Source or Object form.
73
+
74
+ 3. Grant of Patent License. Subject to the terms and conditions of
75
+ this License, each Contributor hereby grants to You a perpetual,
76
+ worldwide, non-exclusive, no-charge, royalty-free, irrevocable
77
+ (except as stated in this section) patent license to make, have made,
78
+ use, offer to sell, sell, import, and otherwise transfer the Work,
79
+ where such license applies only to those patent claims licensable
80
+ by such Contributor that are necessarily infringed by their
81
+ Contribution(s) alone or by combination of their Contribution(s)
82
+ with the Work to which such Contribution(s) was submitted. If You
83
+ institute patent litigation against any entity (including a
84
+ cross-claim or counterclaim in a lawsuit) alleging that the Work
85
+ or a Contribution incorporated within the Work constitutes direct
86
+ or contributory patent infringement, then any patent licenses
87
+ granted to You under this License for that Work shall terminate
88
+ as of the date such litigation is filed.
89
+
90
+ 4. Redistribution. You may reproduce and distribute copies of the
91
+ Work or Derivative Works thereof in any medium, with or without
92
+ modifications, and in Source or Object form, provided that You
93
+ meet the following conditions:
94
+
95
+ (a) You must give any other recipients of the Work or
96
+ Derivative Works a copy of this License; and
97
+
98
+ (b) You must cause any modified files to carry prominent notices
99
+ stating that You changed the files; and
100
+
101
+ (c) You must retain, in the Source form of any Derivative Works
102
+ that You distribute, all copyright, patent, trademark, and
103
+ attribution notices from the Source form of the Work,
104
+ excluding those notices that do not pertain to any part of
105
+ the Derivative Works; and
106
+
107
+ (d) If the Work includes a "NOTICE" text file as part of its
108
+ distribution, then any Derivative Works that You distribute must
109
+ include a readable copy of the attribution notices contained
110
+ within such NOTICE file, excluding those notices that do not
111
+ pertain to any part of the Derivative Works, in at least one
112
+ of the following places: within a NOTICE text file distributed
113
+ as part of the Derivative Works; within the Source form or
114
+ documentation, if provided along with the Derivative Works; or,
115
+ within a display generated by the Derivative Works, if and
116
+ wherever such third-party notices normally appear. The contents
117
+ of the NOTICE file are for informational purposes only and
118
+ do not modify the License. You may add Your own attribution
119
+ notices within Derivative Works that You distribute, alongside
120
+ or as an addendum to the NOTICE text from the Work, provided
121
+ that such additional attribution notices cannot be construed
122
+ as modifying the License.
123
+
124
+ You may add Your own copyright statement to Your modifications and
125
+ may provide additional or different license terms and conditions
126
+ for use, reproduction, or distribution of Your modifications, or
127
+ for any such Derivative Works as a whole, provided Your use,
128
+ reproduction, and distribution of the Work otherwise complies with
129
+ the conditions stated in this License.
130
+
131
+ 5. Submission of Contributions. Unless You explicitly state otherwise,
132
+ any Contribution intentionally submitted for inclusion in the Work
133
+ by You to the Licensor shall be under the terms and conditions of
134
+ this License, without any additional terms or conditions.
135
+ Notwithstanding the above, nothing herein shall supersede or modify
136
+ the terms of any separate license agreement you may have executed
137
+ with Licensor regarding such Contributions.
138
+
139
+ 6. Trademarks. This License does not grant permission to use the trade
140
+ names, trademarks, service marks, or product names of the Licensor,
141
+ except as required for reasonable and customary use in describing the
142
+ origin of the Work and reproducing the content of the NOTICE file.
143
+
144
+ 7. Disclaimer of Warranty. Unless required by applicable law or
145
+ agreed to in writing, Licensor provides the Work (and each
146
+ Contributor provides its Contributions) on an "AS IS" BASIS,
147
+ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
148
+ implied, including, without limitation, any warranties or conditions
149
+ of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
150
+ PARTICULAR PURPOSE. You are solely responsible for determining the
151
+ appropriateness of using or redistributing the Work and assume any
152
+ risks associated with Your exercise of permissions under this License.
153
+
154
+ 8. Limitation of Liability. In no event and under no legal theory,
155
+ whether in tort (including negligence), contract, or otherwise,
156
+ unless required by applicable law (such as deliberate and grossly
157
+ negligent acts) or agreed to in writing, shall any Contributor be
158
+ liable to You for damages, including any direct, indirect, special,
159
+ incidental, or consequential damages of any character arising as a
160
+ result of this License or out of the use or inability to use the
161
+ Work (including but not limited to damages for loss of goodwill,
162
+ work stoppage, computer failure or malfunction, or any and all
163
+ other commercial damages or losses), even if such Contributor
164
+ has been advised of the possibility of such damages.
165
+
166
+ 9. Accepting Warranty or Additional Liability. While redistributing
167
+ the Work or Derivative Works thereof, You may choose to offer,
168
+ and charge a fee for, acceptance of support, warranty, indemnity,
169
+ or other liability obligations and/or rights consistent with this
170
+ License. However, in accepting such obligations, You may act only
171
+ on Your own behalf and on Your sole responsibility, not on behalf
172
+ of any other Contributor, and only if You agree to indemnify,
173
+ defend, and hold each Contributor harmless for any liability
174
+ incurred by, or claims asserted against, such Contributor by reason
175
+ of your accepting any such warranty or additional liability.
176
+
177
+ END OF TERMS AND CONDITIONS
178
+
179
+ APPENDIX: How to apply the Apache License to your work.
180
+
181
+ To apply the Apache License to your work, attach the following
182
+ boilerplate notice, with the fields enclosed by brackets "[]"
183
+ replaced with your own identifying information. (Don't include
184
+ the brackets!) The text should be enclosed in the appropriate
185
+ comment syntax for the file format. We also recommend that a
186
+ file or class name and description of purpose be included on the
187
+ same "printed page" as the copyright notice for easier
188
+ identification within third-party archives.
189
+
190
+ Copyright [yyyy] [name of copyright owner]
191
+
192
+ Licensed under the Apache License, Version 2.0 (the "License");
193
+ you may not use this file except in compliance with the License.
194
+ You may obtain a copy of the License at
195
+
196
+ http://www.apache.org/licenses/LICENSE-2.0
197
+
198
+ Unless required by applicable law or agreed to in writing, software
199
+ distributed under the License is distributed on an "AS IS" BASIS,
200
+ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
201
+ See the License for the specific language governing permissions and
202
+ limitations under the License.
README.md ADDED
@@ -0,0 +1,94 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: apache-2.0
3
+ language:
4
+ - de
5
+ pipeline_tag: text-to-speech
6
+ library_name: onnx
7
+ tags:
8
+ - tts
9
+ - text-to-speech
10
+ - german
11
+ - onnx
12
+ - cpu
13
+ - tiny-voca
14
+ - vocos
15
+ ---
16
+
17
+ # apptor-tiny-tts (German)
18
+
19
+ A tiny German text-to-speech model that runs on the CPU: **4.5M parameters, one 21 MB ONNX file**, about 40x faster
20
+ than real time on four CPU threads. Two voices: `F2` (female) and `M2` (male), 24 kHz.
21
+
22
+ It is built on **[tiny-voca](https://github.com/n4ze3m/tiny-voca) by n4ze3m**: same architecture and training
23
+ recipe, retrained from scratch for German. This is **not an official tiny-voca release**. Thanks to n4ze3m for the
24
+ model and the idea.
25
+
26
+ An optional second stage, a [Vocos](https://github.com/gemelo-ai/vocos) vocoder fine-tuned on the two voices
27
+ (13.5M parameters, 54 MB), re-synthesises the waveform and removes most of the buzzy sound of the tiny built-in
28
+ generator. It is still real time on the CPU (~20x).
29
+
30
+ Listen: `samples/demo.wav` (with Vocos) and `samples/demo-without-vocos.wav`.
31
+
32
+ ## Files
33
+
34
+ | file | what | license |
35
+ |---|---|---|
36
+ | `apptor-tiny-tts-de.onnx` | the TTS model (text tokens -> 24 kHz audio) | Apache-2.0 |
37
+ | `apptor-tiny-tts-de.json` | sample rate, voices, phoneme symbols, limits | Apache-2.0 |
38
+ | `apptor-tiny-tts-vocos-de.pt` | optional: Vocos state dict, fine-tuned from `charactr/vocos-mel-24khz` | MIT |
39
+ | `tinyvoca/` | German text front end (eSpeak NG phonemes, number normalisation), from tiny-voca | Apache-2.0 |
40
+ | `speak.py` | minimal example | Apache-2.0 |
41
+
42
+ ## Use
43
+
44
+ Needs the `espeak-ng` program (`apt install espeak-ng`; GPL-3.0, called as a separate process) and
45
+ `pip install onnxruntime numpy soundfile num2words`.
46
+
47
+ ```bash
48
+ python speak.py "Am 3. Mai um 18:30 Uhr kostet das Ticket 12,50 Euro." --voice M2 --out test.wav
49
+ # with the Vocos second stage (pip install torch vocos):
50
+ python speak.py "Hallo, wie geht es dir?" --voice F2 --vocos --out test.wav
51
+ ```
52
+
53
+ ONNX inputs: `tokens` and `stress` (int64 [1, N], from `tinyvoca.text.phonemize`), `voice` (int64 [1], index into
54
+ `voices`), `speed`, `pitch_shift`, `pitch_range` (float32 [1]; 1.0 / 0.0 / 1.0 for normal). Outputs: `audio`
55
+ (float32 [1, samples]) and `durations`. At most 512 phoneme tokens per call; split longer text into sentences.
56
+
57
+ It is also a voice of [flash-voice](https://github.com/ViktorsGitUser/flash-voice) (`services/tinytts_server.py`,
58
+ same HTTP API as the other voices there), as an Apache-2.0 alternative to the German Phonon model, which is
59
+ CC BY-NC 4.0 (non-commercial only).
60
+
61
+ ## Training
62
+
63
+ - **Data:** about 12 hours of synthetic German speech, ~15,000 utterances, made with
64
+ [Qwen3-TTS](https://github.com/QwenLM/Qwen3-TTS) 1.7B Base (Apache-2.0) with two of its built-in voices. Texts:
65
+ Mozilla Common Voice German sentences (CC0). Utterances were screened with Whisper.
66
+ - **Front end:** eSpeak NG `de`, 67 phoneme symbols (the English inventory of tiny-voca replaced by a German one),
67
+ German number/date/time normalisation.
68
+ - **Model:** tiny-voca recipe, 82k steps on one RTX 3090 (about 10 hours).
69
+ - **Vocoder:** Vocos fine-tuned for about 60k steps on an RTX 5060 Ti: input is tiny-voca's own teacher-forced
70
+ waveform, target the recording.
71
+ - Data, training and tuning took about 12.5 hours on one Sunday.
72
+
73
+ **Evaluation** (Whisper large-v3-turbo, word error rate): 5.0% on 16 unseen sentences (F2 4.6%, M2 5.5%).
74
+
75
+ ## Limitations
76
+
77
+ - German only. English words are read with German pronunciation.
78
+ - The sound is clear but not studio quality; without Vocos it is noticeably buzzy.
79
+ - A word-final long "a" before a comma is sometimes drawn out ("Anna, ...").
80
+ - The voices are synthetic (Qwen3-TTS built-in voices), not clones of real people.
81
+
82
+ ## Deutsch
83
+
84
+ Ein winziges deutsches TTS-Modell für die CPU (4,5M Parameter, 21 MB ONNX), auf Basis von tiny-voca von n4ze3m
85
+ nachtrainiert, mit optionalem Vocos-Vocoder. Gewichte und Code unter Apache 2.0, also auch kommerziell nutzbar,
86
+ der Vocoder unter MIT. Keine offizielle tiny-voca-Version.
87
+
88
+ ## Credits
89
+
90
+ - [tiny-voca](https://github.com/n4ze3m/tiny-voca) by n4ze3m: architecture, training recipe, text front end (Apache-2.0)
91
+ - [Vocos](https://github.com/gemelo-ai/vocos), `charactr/vocos-mel-24khz` (MIT)
92
+ - [Qwen3-TTS](https://github.com/QwenLM/Qwen3-TTS) (Apache-2.0): training speech
93
+ - [Mozilla Common Voice](https://commonvoice.mozilla.org) German (CC0): texts
94
+ - [eSpeak NG](https://github.com/espeak-ng/espeak-ng) (GPL-3.0, not included), num2words (LGPL-2.1)
apptor-tiny-tts-de.json ADDED
@@ -0,0 +1,94 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "sample_rate": 24000,
3
+ "hop": 256,
4
+ "voices": [
5
+ "F2",
6
+ "M2"
7
+ ],
8
+ "symbols": [
9
+ "<pad>",
10
+ "<bos>",
11
+ "<eos>",
12
+ "<sp>",
13
+ "<unk>",
14
+ ",",
15
+ ".",
16
+ "?",
17
+ "!",
18
+ ";",
19
+ ":",
20
+ "\"",
21
+ "'",
22
+ "(",
23
+ ")",
24
+ "—",
25
+ "…",
26
+ "p",
27
+ "b",
28
+ "t",
29
+ "d",
30
+ "k",
31
+ "ɡ",
32
+ "f",
33
+ "v",
34
+ "s",
35
+ "z",
36
+ "ʃ",
37
+ "ʒ",
38
+ "ç",
39
+ "x",
40
+ "h",
41
+ "m",
42
+ "n",
43
+ "ŋ",
44
+ "l",
45
+ "ɾ",
46
+ "r",
47
+ "j",
48
+ "ts",
49
+ "pf",
50
+ "tʃ",
51
+ "dʒ",
52
+ "ə",
53
+ "ɐ",
54
+ "ɜ",
55
+ "ɪ",
56
+ "i",
57
+ "iː",
58
+ "ɛ",
59
+ "ɛː",
60
+ "eː",
61
+ "a",
62
+ "ɑ",
63
+ "ɑː",
64
+ "ɔ",
65
+ "oː",
66
+ "ʊ",
67
+ "uː",
68
+ "y",
69
+ "yː",
70
+ "œ",
71
+ "øː",
72
+ "aɪ",
73
+ "aʊ",
74
+ "ɔø",
75
+ "ʊɐ"
76
+ ],
77
+ "max_tokens": 512,
78
+ "max_frames": 3000,
79
+ "step": 82044,
80
+ "frontend": "espeak",
81
+ "frontend_fingerprint": "ddefc3ebfe4beeca2af195dc920c207110b07bd5067f225c575a3cdeef02d73f",
82
+ "inputs": [
83
+ "tokens",
84
+ "stress",
85
+ "voice",
86
+ "speed",
87
+ "pitch_shift",
88
+ "pitch_range"
89
+ ],
90
+ "outputs": [
91
+ "audio",
92
+ "durations"
93
+ ]
94
+ }
apptor-tiny-tts-de.onnx ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:594d7536b0138957167bd2a6c9e1ca0301f726e981cab14296799444abdb5e46
3
+ size 21379442
apptor-tiny-tts-vocos-de.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5457b535929b1a110eaa49febfb9c383d4fbe948259bee23b33a9db265366679
3
+ size 54369987
samples/demo-without-vocos.wav ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f0f50db9201ff42e40662889ddd1c5e306de3f4690c21a6dcfe6252ca356087c
3
+ size 2148268
samples/demo.wav ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c1cf2afa84ae2fd413aca5b44571d47b4cdf92ed570b70cf4dd8efeca8bce0ac
3
+ size 2148268
speak.py ADDED
@@ -0,0 +1,66 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ """apptor-tiny-tts: German text -> WAV on the CPU.
2
+
3
+ python speak.py "Hallo, wie geht es dir?" --voice F2 --out hallo.wav [--vocos] [--speed 1.0]
4
+ Needs espeak-ng, onnxruntime, numpy, soundfile, num2words; --vocos also torch and vocos.
5
+ """
6
+ import argparse
7
+ import json
8
+ import re
9
+ from pathlib import Path
10
+
11
+ import numpy as np
12
+ import onnxruntime as ort
13
+ import soundfile as sf
14
+
15
+ from tinyvoca.text import phonemize
16
+
17
+ HERE = Path(__file__).resolve().parent
18
+
19
+
20
+ def main():
21
+ ap = argparse.ArgumentParser()
22
+ ap.add_argument("text")
23
+ ap.add_argument("--voice", default="F2", choices=["F2", "M2"])
24
+ ap.add_argument("--out", default="out.wav")
25
+ ap.add_argument("--speed", type=float, default=1.0)
26
+ ap.add_argument("--vocos", action="store_true", help="second stage: fine-tuned Vocos (cleaner sound)")
27
+ ap.add_argument("--threads", type=int, default=4)
28
+ a = ap.parse_args()
29
+
30
+ meta = json.loads((HERE / "apptor-tiny-tts-de.json").read_text(encoding="utf-8"))
31
+ opts = ort.SessionOptions()
32
+ opts.intra_op_num_threads = a.threads
33
+ session = ort.InferenceSession(str(HERE / "apptor-tiny-tts-de.onnx"), opts, providers=["CPUExecutionProvider"])
34
+ voice = meta["voices"].index(a.voice)
35
+
36
+ pieces = []
37
+ for sentence in re.split(r"(?<=[.!?…])\s+", a.text.strip()): # one sentence per call (<= 512 tokens)
38
+ if not sentence:
39
+ continue
40
+ p = phonemize(sentence)
41
+ audio, _ = session.run(None, {
42
+ "tokens": np.array([p.ids], np.int64), "stress": np.array([p.stress], np.int64),
43
+ "voice": np.array([voice], np.int64), "speed": np.array([a.speed], np.float32),
44
+ "pitch_shift": np.array([0.0], np.float32), "pitch_range": np.array([1.0], np.float32)})
45
+ pieces += [audio[0].astype(np.float32), np.zeros(int(0.12 * meta["sample_rate"]), np.float32)]
46
+ audio = np.concatenate(pieces)
47
+
48
+ if a.vocos:
49
+ import torch
50
+ from vocos import Vocos
51
+ torch.set_num_threads(a.threads)
52
+ voc = Vocos.from_pretrained("charactr/vocos-mel-24khz").eval()
53
+ voc.load_state_dict(torch.load(HERE / "apptor-tiny-tts-vocos-de.pt", map_location="cpu", weights_only=True))
54
+ with torch.inference_mode():
55
+ x = torch.from_numpy(audio)[None]
56
+ audio = voc.decode(voc.feature_extractor(x))[0, :len(audio)].numpy()
57
+
58
+ peak = float(np.abs(audio).max())
59
+ if peak > 0.95:
60
+ audio = audio * (0.95 / peak)
61
+ sf.write(a.out, audio, meta["sample_rate"])
62
+ print(f"wrote {a.out} ({len(audio) / meta['sample_rate']:.1f} s)")
63
+
64
+
65
+ if __name__ == "__main__":
66
+ main()
tinyvoca/__init__.py ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ """Text front end of tiny-voca (https://github.com/n4ze3m/tiny-voca, Apache 2.0), German fork:
2
+ text.py (eSpeak NG "de", German phone inventory) and normalize_de.py, copied unchanged from the German training
3
+ code so the server makes exactly the tokens the model was trained on. Needs the espeak-ng program and num2words."""
tinyvoca/normalize_de.py ADDED
@@ -0,0 +1,65 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ """German text normalization before eSpeak: spell out what a speaker reads aloud
2
+ differently from eSpeak's defaults (abbreviations, clock times, money, ordinals
3
+ before months, percent, years). Used for annotation, training and inference alike.
4
+ """
5
+ import re
6
+
7
+ try:
8
+ from num2words import num2words as _n2w
9
+ except ImportError: # eSpeak still reads plain numbers; only the special forms need num2words
10
+ _n2w = None
11
+
12
+ MONTHS = ("Januar", "Februar", "März", "April", "Mai", "Juni", "Juli", "August", "September", "Oktober", "November", "Dezember")
13
+ ABBREVIATIONS = {
14
+ "z.B.": "zum Beispiel", "z. B.": "zum Beispiel", "d.h.": "das heißt", "d. h.": "das heißt", "u.a.": "unter anderem",
15
+ "usw.": "und so weiter", "bzw.": "beziehungsweise", "ca.": "circa", "vgl.": "vergleiche", "evtl.": "eventuell",
16
+ "ggf.": "gegebenenfalls", "inkl.": "inklusive", "Nr.": "Nummer", "Dr.": "Doktor", "Prof.": "Professor",
17
+ "Str.": "Straße", "Hr.": "Herr", "Fr.": "Frau", "St.": "Sankt", "etc.": "et cetera", "Mio.": "Millionen",
18
+ "Mrd.": "Milliarden", "km/h": "Kilometer pro Stunde", "€": " Euro", "%": " Prozent",
19
+ }
20
+
21
+
22
+ def _num(value, to="cardinal"):
23
+ if _n2w is None:
24
+ return str(value)
25
+ return _n2w(int(value), lang="de", to=to)
26
+
27
+
28
+ def _ordinal(match):
29
+ word = _num(match[2], "ordinal") # "dritte"
30
+ prefix = match[1] or ""
31
+ # After a preposition/article in the dative ("am", "vom", "zum", "bis zum") the ending is -n.
32
+ if prefix.strip().lower() in ("am", "vom", "zum", "im", "ab dem", "bis zum", "seit dem", "dem", "den"):
33
+ word += "n"
34
+ return f"{prefix}{word} {match[3]}"
35
+
36
+
37
+ def _time(match):
38
+ hours, minutes = int(match[1]), int(match[2])
39
+ text = f"{_num(hours)} Uhr"
40
+ if hours == 1:
41
+ text = "ein Uhr"
42
+ return text + (f" {_num(minutes)}" if minutes else "")
43
+
44
+
45
+ def _money(match):
46
+ euros, cents = int(match[1]), int(match[2])
47
+ text = f"{_num(euros)} Euro"
48
+ return text + (f" {_num(cents)}" if cents else "")
49
+
50
+
51
+ def normalize(text):
52
+ for short, long in ABBREVIATIONS.items():
53
+ text = text.replace(short, long)
54
+ text = re.sub(r"\b(\d{1,2}):(\d{2})\s*Uhr\b", _time, text)
55
+ text = re.sub(r"\b(\d{1,2}):(\d{2})\b", _time, text)
56
+ text = re.sub(r"\b(\d+),(\d{2})\s*Euro\b", _money, text)
57
+ text = re.sub(r"\b((?:[Aa]m|[Vv]om|[Zz]um|[Bb]is zum|[Ss]eit dem|[Aa]b dem) )?(\d{1,2})\. (" + "|".join(MONTHS) + r")\b", _ordinal, text)
58
+ # Years after "im Jahr", or bare four-digit numbers 1100-1999 in running text, are read as years.
59
+ text = re.sub(r"\b(1[1-9]\d\d)\b", lambda m: _num(m[1], "year"), text)
60
+ text = re.sub(r"(\d+),(\d+)", lambda m: f"{_num(m[1])} Komma {' '.join(_num(d) for d in m[2])}", text)
61
+ text = re.sub(r"(?<![\w-])-(\d+)", lambda m: "minus " + m[1], text)
62
+ if _n2w is not None:
63
+ text = re.sub(r"\d+", lambda m: _num(m[0]), text)
64
+ text = re.sub(r"\bein(hundert|tausend)", r"\1", text)
65
+ return re.sub(r"\s+", " ", text).strip()
tinyvoca/text.py ADDED
@@ -0,0 +1,281 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ """German front end (fork of the English tiny-voca front end): text -> phoneme tokens with a stress feature.
2
+
3
+ The same code runs during annotation, training-set preparation and inference,
4
+ so tokens can never drift between them. Two programs can produce the phonemes:
5
+
6
+ * ``espeak`` (front-end version 3, the default): eSpeak NG, as described below.
7
+ The current checkpoints and annotations were made with it.
8
+ * ``own`` (front-end version 4): the separate ``dizione`` package, a dictionary
9
+ built from CMUdict, a small model for unknown words and rules. No eSpeak, so
10
+ the whole system can be shipped under a permissive licence. Same 76 symbols.
11
+
12
+ Choose with ``set_frontend("own")`` or ``TINY_VOCA_FRONTEND=own``. Everything
13
+ below about tokens holds for both; the rest of this note is the eSpeak path.
14
+
15
+ * eSpeak NG (``en-us``) is the only pronunciation source. It is an external
16
+ rule-based program: no neural G2P and no downloaded weights.
17
+ * A token is one eSpeak phoneme exactly as eSpeak segments it (``--sep``), for
18
+ example ``t``, ``aɪ``, ``ɑːɹ`` or ``n̩``. This is also the label inventory of
19
+ the phoneme recogniser used for offline forced alignment, so one aligned
20
+ label corresponds to exactly one token.
21
+ * Lexical stress is a parallel feature (0 none, 1 primary, 2 secondary), not a
22
+ token, so it never has to be given audio frames.
23
+ * Punctuation, word boundaries and the utterance edges are explicit tokens that
24
+ may last zero frames. They carry pauses when the recording has them.
25
+
26
+ Versions 1 and 2 encoded IPA as UTF-8 bytes and piped text to eSpeak without a
27
+ final newline; eSpeak NG 1.51 then drops the last character of the input, so
28
+ the final phonemes of every clause were wrong. Version 3 terminates the input
29
+ and has a regression test for it.
30
+ """
31
+
32
+ from dataclasses import dataclass
33
+ import hashlib
34
+ import json
35
+ import os
36
+ import re
37
+ import shutil
38
+ import subprocess
39
+ import unicodedata
40
+
41
+ FRONTEND_VERSION = 103 # German fork: eSpeak NG "de", German phone inventory
42
+ ESPEAK_VOICE = "de"
43
+ SEPARATOR = "_"
44
+ FRONTENDS = ("espeak", "own")
45
+ FRONTEND_VARIABLE = "TINY_VOCA_FRONTEND"
46
+ _frontend = os.environ.get(FRONTEND_VARIABLE, "espeak")
47
+ if _frontend not in FRONTENDS:
48
+ raise ValueError(f"{FRONTEND_VARIABLE} must be one of {FRONTENDS}, not {_frontend!r}")
49
+
50
+ PAD, BOS, EOS, SP, UNK = "<pad>", "<bos>", "<eos>", "<sp>", "<unk>"
51
+ SPECIALS = (PAD, BOS, EOS, SP, UNK)
52
+ PUNCTUATION = (",", ".", "?", "!", ";", ":", '"', "'", "(", ")", "—", "…")
53
+ # German inventory: every phone is a label of the forced-alignment recogniser, except
54
+ # "ʊɐ", which eSpeak NG prints (and the recogniser was trained on) as "??".
55
+ CONSONANTS = ("p", "b", "t", "d", "k", "ɡ", "f", "v", "s", "z", "ʃ", "ʒ", "ç", "x", "h",
56
+ "m", "n", "ŋ", "l", "ɾ", "r", "j", "ts", "pf", "tʃ", "dʒ")
57
+ VOWELS = ("ə", "ɐ", "ɜ", "ɪ", "i", "iː", "ɛ", "ɛː", "eː", "a", "ɑ", "ɑː", "ɔ", "oː", "ʊ", "uː",
58
+ "y", "yː", "œ", "øː", "aɪ", "aʊ", "ɔø", "ʊɐ")
59
+ PHONES = CONSONANTS + VOWELS
60
+ SYMBOLS = SPECIALS + PUNCTUATION + PHONES
61
+ SYMBOL_TO_ID = {symbol: index for index, symbol in enumerate(SYMBOLS)}
62
+ PHONE_SET = frozenset(PHONES)
63
+ PHONE_IDS = frozenset(SYMBOL_TO_ID[p] for p in PHONES) | {SYMBOL_TO_ID[UNK]}
64
+ VOICELESS = frozenset(("p", "t", "k", "f", "s", "ʃ", "ç", "x", "h", "ts", "pf", "tʃ"))
65
+ VOWEL_SET = frozenset(VOWELS)
66
+ STRESS_MARKS = {"ˈ": 1, "ˌ": 2}
67
+
68
+ # eSpeak phonemes outside the inventory (English/French loan words, rare allophones).
69
+ FALLBACK = {
70
+ "??": ("ʊɐ",), "ɹ": ("ɾ",), "ʁ": ("ɾ",), "w": ("v",), "ð": ("d",), "θ": ("s",), "ʔ": (),
71
+ "e": ("eː",), "o": ("oː",), "u": ("uː",), "ø": ("øː",), "ɒ": ("ɔ",), "ʌ": ("a",), "æ": ("ɛ",),
72
+ "ɜː": ("øː",), "ɔː": ("oː",), "ɪː": ("iː",), "aː": ("ɑː",), "eɪ": ("eː",), "ɛɪ": ("eː",),
73
+ "əʊ": ("oː",), "oʊ": ("oː",), "eə": ("ɛː",), "ɪə": ("iː", "ɐ"), "ʊə": ("uː", "ɐ"), "aɪə": ("aɪ", "ɐ"),
74
+ "ɔɪ": ("ɔø",), "ɔʏ": ("ɔø",), "əl": ("ə", "l"), "l̩": ("ə", "l"), "n̩": ("ə", "n"), "m̩": ("ə", "m"),
75
+ "ɑ̃": ("ɑ", "ŋ"), "ɔ̃": ("ɔ", "ŋ"), "ɛ̃": ("ɛ", "ŋ"), "œ̃": ("œ", "ŋ"), "ʏ": ("y",), "ɚ": ("ɐ",),
76
+ "ɲ": ("n", "j"), "ɬ": ("l",), "ɫ": ("l",), "dz": ("d", "z"), "β": ("v",), "ɣ": ("ɡ",), "ʎ": ("l", "j"),
77
+ }
78
+
79
+ _QUOTES = str.maketrans({"‘": "'", "’": "'", "‚": "'", "‛": "'", "`": "'", "´": "'",
80
+ "“": '"', "”": '"', "„": '"', "‟": '"', "«": '"', "»": '"',
81
+ "–": "—", "―": "—", "[": "(", "{": "(", "]": ")", "}": ")"})
82
+ _PUNCTUATION_CHARS = frozenset(PUNCTUATION)
83
+ ABBREVIATIONS = {"dr", "prof", "nr", "str", "bzw", "ca", "usw", "vgl", "evtl", "ggf", "inkl", "hr", "fr", "st", "z", "b", "d", "h", "u", "a", "s", "o"}
84
+ _LANGUAGE_TAG = re.compile(r"\((?:_?[a-z0-9]_?){2,3}(?:-(?:_?[a-z0-9]_?)+)*\)")
85
+
86
+
87
+ @dataclass(frozen=True)
88
+ class Phonemized:
89
+ symbols: tuple
90
+ ids: tuple
91
+ stress: tuple
92
+ unknown: tuple # eSpeak phonemes that had no inventory mapping
93
+
94
+ def __len__(self):
95
+ return len(self.ids)
96
+
97
+
98
+ def frontend():
99
+ """Name of the program that produces the phonemes: ``"espeak"`` or ``"own"``."""
100
+ return _frontend
101
+
102
+
103
+ def set_frontend(name):
104
+ """Select the phonemizer for this process. Tokens of the two front ends
105
+ must not be mixed: annotations and checkpoints record which one made them."""
106
+ global _frontend
107
+ if name not in FRONTENDS:
108
+ raise ValueError(f"front end must be one of {FRONTENDS}, not {name!r}")
109
+ _frontend = name
110
+
111
+
112
+ def _dizione():
113
+ try:
114
+ import dizione
115
+ except ImportError as error:
116
+ raise RuntimeError('the "own" front end needs the dizione package: pip install dizione') from error
117
+ if tuple(dizione.symbols.SYMBOLS) != SYMBOLS:
118
+ raise RuntimeError("dizione's symbol inventory differs from this model's")
119
+ return dizione
120
+
121
+
122
+ def frontend_identity():
123
+ if _frontend == "own":
124
+ phonemizer = _dizione()
125
+ return {"frontend_version": phonemizer.VERSION, "phonemizer": "dizione", "symbols": list(SYMBOLS)}
126
+ return {"frontend_version": FRONTEND_VERSION, "voice": ESPEAK_VOICE, "separator": SEPARATOR,
127
+ "symbols": list(SYMBOLS), "fallback": {key: list(value) for key, value in sorted(FALLBACK.items())}}
128
+
129
+
130
+ def frontend_runtime():
131
+ """What an annotation run records about the exact program behind its
132
+ tokens: the eSpeak version, or the fingerprint of the phonemizer build."""
133
+ if _frontend == "own":
134
+ return {"phonemizer": _dizione().identity()["fingerprint"]}
135
+ return {"espeak_version": espeak_version()}
136
+
137
+
138
+ def frontend_fingerprint():
139
+ return hashlib.sha256(json.dumps(frontend_identity(), sort_keys=True, ensure_ascii=False).encode()).hexdigest()
140
+
141
+
142
+ def normalize_text(text):
143
+ text = unicodedata.normalize("NFKC", text).translate(_QUOTES)
144
+ text = re.sub(r"\.{3,}", "…", text)
145
+ text = re.sub(r"-{2,}|\s-\s", "—", text)
146
+ return re.sub(r"\s+", " ", text).strip()
147
+
148
+
149
+ def chunks(text):
150
+ """Yield ``(is_punctuation, piece)``. Decimals, clock times, digit groups,
151
+ contractions, title abbreviations and initialisms stay inside the text piece
152
+ for eSpeak; every other punctuation mark becomes its own token."""
153
+ start = 0
154
+ for i, char in enumerate(text):
155
+ if char not in _PUNCTUATION_CHARS:
156
+ continue
157
+ previous = text[i-1] if i else ""
158
+ following = text[i+1] if i+1 < len(text) else ""
159
+ if char == "'" and previous.isalnum() and following.isalnum():
160
+ continue
161
+ if char in ".,:" and previous.isdigit() and following.isdigit():
162
+ continue
163
+ if char == "." and previous.isdigit() and following == " " and i+2 < len(text) and text[i+2].isalnum():
164
+ continue
165
+ if char == ".":
166
+ word = re.search(r"([A-Za-z]+)$", text[:i])
167
+ if word and word[1].lower() in ABBREVIATIONS:
168
+ continue
169
+ if previous.isalpha() and following.isalpha():
170
+ continue
171
+ if text[start:i].strip():
172
+ yield False, text[start:i].strip()
173
+ yield True, char
174
+ start = i+1
175
+ if text[start:].strip():
176
+ yield False, text[start:].strip()
177
+
178
+
179
+ def espeak_executable():
180
+ executable = shutil.which("espeak-ng")
181
+ if executable is None:
182
+ raise RuntimeError("Install espeak-ng on the cloud machine before preprocessing/inference")
183
+ return executable
184
+
185
+
186
+ def espeak_version():
187
+ return subprocess.run([espeak_executable(), "--version"], capture_output=True, text=True, check=True, timeout=10).stdout.strip()
188
+
189
+
190
+ def espeak_words(piece):
191
+ """Phonemize one text piece. Returns words as lists of ``(phone, stress)``.
192
+
193
+ The input must end with a newline: eSpeak NG 1.51 silently drops the last
194
+ character of unterminated standard input.
195
+ """
196
+ result = subprocess.run(
197
+ [espeak_executable(), "-q", "--ipa", "--sep=" + SEPARATOR, "-v", ESPEAK_VOICE, "--stdin"],
198
+ input=piece.rstrip("\n") + "\n", text=True, capture_output=True, check=True, timeout=30)
199
+ return parse_espeak(result.stdout)
200
+
201
+
202
+ def parse_espeak(output):
203
+ words = []
204
+ for word in _LANGUAGE_TAG.sub("", output).split():
205
+ units = []
206
+ for unit in word.split(SEPARATOR):
207
+ stress = 0
208
+ for mark, level in STRESS_MARKS.items():
209
+ if mark in unit:
210
+ stress = stress or level
211
+ unit = unit.replace(mark, "")
212
+ unit = "".join(c for c in unit if unicodedata.category(c) != "Cf" and c not in "()")
213
+ if unit:
214
+ units.append((unit, stress))
215
+ if units:
216
+ words.append(units)
217
+ return words
218
+
219
+
220
+ def canonical(phone):
221
+ """Map an eSpeak phoneme to inventory phones; ``None`` when unknown."""
222
+ if phone in PHONE_SET:
223
+ return (phone,)
224
+ if phone in FALLBACK:
225
+ return FALLBACK[phone]
226
+ stripped = "".join(c for c in unicodedata.normalize("NFD", phone) if not unicodedata.combining(c))
227
+ for candidate in (stripped, stripped.replace("ː", ""), stripped + "ː"):
228
+ if candidate != phone and candidate in PHONE_SET:
229
+ return (candidate,)
230
+ if candidate != phone and candidate in FALLBACK:
231
+ return FALLBACK[candidate]
232
+ return None
233
+
234
+
235
+ def tokenize(pieces):
236
+ """Build the token sequence from ``(is_punctuation, value)`` pieces where a
237
+ text piece's value is the parsed eSpeak word list. Pure function (testable
238
+ without eSpeak)."""
239
+ symbols, stress, unknown = [BOS], [0], []
240
+ for punctuation, value in pieces:
241
+ if punctuation:
242
+ if symbols[-1] != value: # collapse repeats such as "!!"
243
+ symbols.append(value)
244
+ stress.append(0)
245
+ continue
246
+ for word in value:
247
+ if symbols[-1] in PHONE_SET or symbols[-1] == UNK:
248
+ symbols.append(SP)
249
+ stress.append(0)
250
+ for phone, level in word:
251
+ mapped = canonical(phone)
252
+ if mapped is None:
253
+ unknown.append(phone)
254
+ mapped = (UNK,)
255
+ # A fallback may expand to several phones: stress goes to its first vowel.
256
+ carrier = next((i for i, item in enumerate(mapped) if item in VOWEL_SET), 0)
257
+ for index, item in enumerate(mapped):
258
+ symbols.append(item)
259
+ stress.append(level if index == carrier else 0)
260
+ symbols.append(EOS)
261
+ stress.append(0)
262
+ if not any(s in PHONE_SET or s == UNK for s in symbols):
263
+ raise ValueError("the text has nothing to pronounce")
264
+ return Phonemized(tuple(symbols), tuple(SYMBOL_TO_ID[s] for s in symbols), tuple(stress), tuple(unknown))
265
+
266
+
267
+ def phonemize(text):
268
+ if not isinstance(text, str) or not text.strip():
269
+ raise ValueError("text must not be empty")
270
+ if _frontend == "own":
271
+ return tokenize(_dizione().pieces(text))
272
+ pieces = []
273
+ from .normalize_de import normalize as normalize_german
274
+ for punctuation, piece in chunks(normalize_text(normalize_german(normalize_text(text)))):
275
+ pieces.append((True, piece) if punctuation else (False, espeak_words(piece)))
276
+ return tokenize(pieces)
277
+
278
+
279
+ def is_phone(token_ids):
280
+ """Boolean list/tensor helper: tokens that must be realised as audio."""
281
+ return [int(i) in PHONE_IDS for i in token_ids]