spybyscript commited on
Commit
497346d
·
verified ·
1 Parent(s): 24f9bf8

Publish inference weights only and add third-party notices

Browse files
README.md CHANGED
@@ -94,7 +94,7 @@ All ten have Spanish (`es`), Italian (`it`), French (`fr`), German (`de`) and Vi
94
  - `coreai/` (155 MB): Core AI bundle for apps on iOS, iPadOS and visionOS 27 (`.aimodel` assets); it also runs on macOS 27, where the runtime uses the Core ML bundle by default. INT8 weights per output channel with FP16 compute; size buckets are functions of one asset sharing weights. The flow also runs accurately on the Neural Engine (`controls.coreai.neural_engine_assets`), which is the placement for apps that speak in the background; the GPU is the default.
95
  - `coreml/` (142 MB): Core ML bundle for Apple silicon with iOS 18 / macOS 15 and later (compiled `.mlmodelc` assets), with INT8 weights and FP16 compute; GPU by default, CPU in background apps.
96
  - `onnx/`: ONNX Runtime bundle with dynamic INT8 transformer weights; the vocoder stays in full precision.
97
- - `pytorch/`: duration, vector, adapter/vocoder, autoencoder and G2P checkpoints, with the release configuration assets.
98
  - `SHA256SUMS.json`: published-file hashes.
99
 
100
  Every runtime bundle includes the G2P model in a matching precision. The Apple bundles also give the G2P narrower input widths (64–512 tokens), so short phrases are converted faster.
@@ -116,4 +116,4 @@ The system supports managed voices, not arbitrary speaker cloning. Intended uses
116
 
117
  ## License and acknowledgments
118
 
119
- Apache 2.0. LiteRT and ONNX Runtime provide inference; Vocos (`charactr/vocos-mel-24khz`) is the base of the 24 kHz waveform decoder. DeepPhonemizer-derived tooling, Wiktionary-derived vocabulary, eSpeak phonemization and Montreal Forced Aligner contributed to the frontend and alignment workflow.
 
94
  - `coreai/` (155 MB): Core AI bundle for apps on iOS, iPadOS and visionOS 27 (`.aimodel` assets); it also runs on macOS 27, where the runtime uses the Core ML bundle by default. INT8 weights per output channel with FP16 compute; size buckets are functions of one asset sharing weights. The flow also runs accurately on the Neural Engine (`controls.coreai.neural_engine_assets`), which is the placement for apps that speak in the background; the GPU is the default.
95
  - `coreml/` (142 MB): Core ML bundle for Apple silicon with iOS 18 / macOS 15 and later (compiled `.mlmodelc` assets), with INT8 weights and FP16 compute; GPU by default, CPU in background apps.
96
  - `onnx/`: ONNX Runtime bundle with dynamic INT8 transformer weights; the vocoder stays in full precision.
97
+ - `pytorch/`: duration, vector, adapter/vocoder and G2P weights (no training state), with the release configuration assets.
98
  - `SHA256SUMS.json`: published-file hashes.
99
 
100
  Every runtime bundle includes the G2P model in a matching precision. The Apple bundles also give the G2P narrower input widths (64–512 tokens), so short phrases are converted faster.
 
116
 
117
  ## License and acknowledgments
118
 
119
+ Apache 2.0. LiteRT and ONNX Runtime provide inference; Vocos (`charactr/vocos-mel-24khz`) is the base of the 24 kHz waveform decoder. DeepPhonemizer-derived tooling, Wiktionary-derived vocabulary, eSpeak phonemization and Montreal Forced Aligner contributed to the frontend and alignment workflow. Vocos and DeepPhonemizer are MIT-licensed; their notices are in `THIRD_PARTY_NOTICES.md`.
SHA256SUMS.json CHANGED
@@ -4,8 +4,12 @@
4
  "bytes": 11357
5
  },
6
  "README.md": {
7
- "sha256": "c370fead2b33fee0a90b460b9eced30213bdbb3f3cc36e5439a343e63962eb59",
8
- "bytes": 9498
 
 
 
 
9
  },
10
  "coreai/assets/fixed_references.json": {
11
  "sha256": "d3c26e7e90d90fa719254ff1ff05dc4b391cd47231b7556c3c2ed5f63547925b",
@@ -419,10 +423,6 @@
419
  "sha256": "4843d57a422b0a6045bbc6fc9a99afa329126a6742ff95ebcefcda660f3ee474",
420
  "bytes": 7769
421
  },
422
- "pytorch/autoencoder.pt": {
423
- "sha256": "09b837f52b5faa572d5cfff50476d9c9c449dd3b33dacbc21c95d4cbcf76d10d",
424
- "bytes": 176762621
425
- },
426
  "pytorch/condition_index.json": {
427
  "sha256": "995750fb8ff573f087f75df851fd38dd7d574e3a47773e8f150bb3b928cf251f",
428
  "bytes": 612
@@ -436,16 +436,16 @@
436
  "bytes": 786473
437
  },
438
  "pytorch/g2p.pt": {
439
- "sha256": "96572c6f7d232b1cd229bf50aebd92b74d7555f35df223d29c860199e900ea31",
440
- "bytes": 332035717
441
  },
442
  "pytorch/phone_vocab.json": {
443
  "sha256": "83e6457a69948edbd9a94b404c22fd408c263eb0d0b5acaafa4926e994587721",
444
  "bytes": 1793
445
  },
446
  "pytorch/release_config.json": {
447
- "sha256": "f952032e5fa73181c7b20a29d15c4ef1fd7b59516b45f5bf6bcefa6282abdc22",
448
- "bytes": 931
449
  },
450
  "pytorch/vector_estimator.pt": {
451
  "sha256": "b5e0e57e4aae30d7d877121fa8b526f3f5105e5e97d5a0bab55e4bcba2a67b49",
 
4
  "bytes": 11357
5
  },
6
  "README.md": {
7
+ "sha256": "4f2d6c847adb1718666fca546eb0f6730b4177a2b7ca117aa165a2a83ca3b572",
8
+ "bytes": 9591
9
+ },
10
+ "THIRD_PARTY_NOTICES.md": {
11
+ "sha256": "3b793b8ee250cc711e4012ee42e9625a8a7e788bf12cab85a1abbd17b17644ca",
12
+ "bytes": 2818
13
  },
14
  "coreai/assets/fixed_references.json": {
15
  "sha256": "d3c26e7e90d90fa719254ff1ff05dc4b391cd47231b7556c3c2ed5f63547925b",
 
423
  "sha256": "4843d57a422b0a6045bbc6fc9a99afa329126a6742ff95ebcefcda660f3ee474",
424
  "bytes": 7769
425
  },
 
 
 
 
426
  "pytorch/condition_index.json": {
427
  "sha256": "995750fb8ff573f087f75df851fd38dd7d574e3a47773e8f150bb3b928cf251f",
428
  "bytes": 612
 
436
  "bytes": 786473
437
  },
438
  "pytorch/g2p.pt": {
439
+ "sha256": "0085f558837c85cdf570337f7af37742b6158551a2c7f5078edcd35c6e84f7db",
440
+ "bytes": 119207485
441
  },
442
  "pytorch/phone_vocab.json": {
443
  "sha256": "83e6457a69948edbd9a94b404c22fd408c263eb0d0b5acaafa4926e994587721",
444
  "bytes": 1793
445
  },
446
  "pytorch/release_config.json": {
447
+ "sha256": "08612d7403dcb42e5061508acf59cd16e9ac1855d69a1b665fcf388ba01d84fb",
448
+ "bytes": 841
449
  },
450
  "pytorch/vector_estimator.pt": {
451
  "sha256": "b5e0e57e4aae30d7d877121fa8b526f3f5105e5e97d5a0bab55e4bcba2a67b49",
THIRD_PARTY_NOTICES.md ADDED
@@ -0,0 +1,62 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Third-party notices
2
+
3
+ Scylla's Band is released under the Apache License 2.0 (see `LICENSE`). Parts of the models are derived from, or
4
+ store objects defined by, the following MIT-licensed projects.
5
+
6
+ ## Vocos
7
+
8
+ The 24 kHz waveform decoder in every bundle (`litert/`, `onnx/`, `coreml/`, `coreai/`) and in `pytorch/vocoder.pt` is
9
+ fine-tuned from the Vocos model `charactr/vocos-mel-24khz` (https://github.com/gemelo-ai/vocos).
10
+
11
+ ```
12
+ MIT License
13
+
14
+ Copyright (c) 2023 Charactr Inc.
15
+
16
+ Permission is hereby granted, free of charge, to any person obtaining a copy
17
+ of this software and associated documentation files (the "Software"), to deal
18
+ in the Software without restriction, including without limitation the rights
19
+ to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
20
+ copies of the Software, and to permit persons to whom the Software is
21
+ furnished to do so, subject to the following conditions:
22
+
23
+ The above copyright notice and this permission notice shall be included in all
24
+ copies or substantial portions of the Software.
25
+
26
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
27
+ IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
28
+ FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
29
+ AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
30
+ LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
31
+ OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
32
+ SOFTWARE.
33
+ ```
34
+
35
+ ## DeepPhonemizer
36
+
37
+ The G2P model uses the DeepPhonemizer architecture, and `pytorch/g2p.pt` stores DeepPhonemizer preprocessing objects
38
+ (https://github.com/as-ideas/DeepPhonemizer).
39
+
40
+ ```
41
+ MIT License
42
+
43
+ Copyright (c) 2021 Axel Springer News Media & Tech GmbH & Co. KG - Ideas Engineering
44
+
45
+ Permission is hereby granted, free of charge, to any person obtaining a copy
46
+ of this software and associated documentation files (the "Software"), to deal
47
+ in the Software without restriction, including without limitation the rights
48
+ to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
49
+ copies of the Software, and to permit persons to whom the Software is
50
+ furnished to do so, subject to the following conditions:
51
+
52
+ The above copyright notice and this permission notice shall be included in all
53
+ copies or substantial portions of the Software.
54
+
55
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
56
+ IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
57
+ FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
58
+ AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
59
+ LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
60
+ OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
61
+ SOFTWARE.
62
+ ```
pytorch/autoencoder.pt DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:09b837f52b5faa572d5cfff50476d9c9c449dd3b33dacbc21c95d4cbcf76d10d
3
- size 176762621
 
 
 
 
pytorch/g2p.pt CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:96572c6f7d232b1cd229bf50aebd92b74d7555f35df223d29c860199e900ea31
3
- size 332035717
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0085f558837c85cdf570337f7af37742b6158551a2c7f5078edcd35c6e84f7db
3
+ size 119207485
pytorch/release_config.json CHANGED
@@ -1,11 +1,10 @@
1
  {
2
  "schema": "scyllasband_pytorch_release_v2",
3
  "files": {
4
- "autoencoder.pt": "09b837f52b5faa572d5cfff50476d9c9c449dd3b33dacbc21c95d4cbcf76d10d",
5
  "condition_index.json": "995750fb8ff573f087f75df851fd38dd7d574e3a47773e8f150bb3b928cf251f",
6
  "duration.pt": "765e8970c5c014be982a0645129682b47338656ec407373683da78318a77a5dd",
7
  "fixed_references.json": "0120e1cf1ef9b3ee19ed3d269c4dc6f6c47e694a59bcce5bba89abe101cd20e4",
8
- "g2p.pt": "96572c6f7d232b1cd229bf50aebd92b74d7555f35df223d29c860199e900ea31",
9
  "phone_vocab.json": "83e6457a69948edbd9a94b404c22fd408c263eb0d0b5acaafa4926e994587721",
10
  "vector_estimator.pt": "b5e0e57e4aae30d7d877121fa8b526f3f5105e5e97d5a0bab55e4bcba2a67b49",
11
  "vocoder.pt": "e31e681d8c524f149efc2a0ea55474c28c1e1c44a5188378ba9d14918db2a15d"
 
1
  {
2
  "schema": "scyllasband_pytorch_release_v2",
3
  "files": {
 
4
  "condition_index.json": "995750fb8ff573f087f75df851fd38dd7d574e3a47773e8f150bb3b928cf251f",
5
  "duration.pt": "765e8970c5c014be982a0645129682b47338656ec407373683da78318a77a5dd",
6
  "fixed_references.json": "0120e1cf1ef9b3ee19ed3d269c4dc6f6c47e694a59bcce5bba89abe101cd20e4",
7
+ "g2p.pt": "0085f558837c85cdf570337f7af37742b6158551a2c7f5078edcd35c6e84f7db",
8
  "phone_vocab.json": "83e6457a69948edbd9a94b404c22fd408c263eb0d0b5acaafa4926e994587721",
9
  "vector_estimator.pt": "b5e0e57e4aae30d7d877121fa8b526f3f5105e5e97d5a0bab55e4bcba2a67b49",
10
  "vocoder.pt": "e31e681d8c524f149efc2a0ea55474c28c1e1c44a5188378ba9d14918db2a15d"