993da778b4
chore: use original instead of scarf urls
...
These allowed Coqui to get download stats, which we don't need anymore
2024-11-10 22:19:25 +01:00
2df9bfa78e
refactor: handle deprecation of torch.cuda.amp.autocast ( #144 )
...
torch.cuda.amp.autocast(args...) and torch.cpu.amp.autocast(args...) will be
deprecated. Please use torch.autocast("cuda", args...) or torch.autocast("cpu",
args...) instead.
https://pytorch.org/docs/stable/amp.html
2024-11-09 18:37:08 +01:00
540e8d6cf2
fix(bin.synthesize): return speakers names only ( #147 )
2024-11-09 18:35:54 +01:00
0971bc236e
refactor: use external package for monotonic alignment
2024-11-08 12:35:04 +01:00
6314032fd7
Merge pull request #113 from idiap/pytorch
...
fix: only enable load with weights_only in pytorch>=2.4
2024-11-04 22:14:42 +01:00
8e66be2c32
fix: only enable load with weights_only in pytorch>=2.4
...
Allows moving the minimum Pytorch version back to 2.1
2024-11-04 18:39:51 +01:00
47ad0bf190
fix(text.characters): add nasal diacritic ( #127 )
2024-11-04 17:40:23 +01:00
964b813235
fix(gpt): set attention mask and address other warnings
2024-10-25 18:49:04 +02:00
018d4ba1db
fix(xtts): support transformers>=4.43.0 in streaming inference
2024-10-20 15:20:26 +02:00
36611a7192
feat: normalize unicode characters in text cleaners ( #85 )
...
* Add normalizer type C to text cleaners
* Linter recommendations
* Add unicode normalize to every cleaner
* Format test_text_cleaners.py
2024-10-02 17:01:19 +02:00
e5dd06b3bb
Merge pull request #77 from shavit/71-torch-load
...
Load weights only in torch.load
2024-09-12 23:28:57 +01:00
659b4852ba
chore(bark): remove manual download of hubert model
...
Bark was previously adapted to download Hubert from HuggingFace, so the manual
download is superfluous.
2024-09-12 23:37:19 +02:00
86b58fb6d9
fix: define torch safe globals for torch.load
...
Required for loading some models using torch.load(..., weights_only=True). This
is only available from Pytorch 2.4
2024-09-12 23:37:19 +02:00
17ca24c3d6
fix: load weights only in torch.load
2024-09-12 23:37:19 +02:00
1920328822
feat(xtts): support hindi in tokenizer ( #64 )
...
Added proper tokenizer support for Hindi Language which would prevent crash while fine tuning Hindi language.
Co-authored-by: Akshat Bhardwaj <157223825+akshatrocky@users.noreply.github.com>
2024-09-12 21:29:21 +02:00
9c604c1de0
chore(dataset): address lint issues
2024-07-31 15:47:27 +02:00
8c460d0cd0
fix(dataset): skip files where audio length can't be computed
...
Avoids hard failures when the audio can't be decoded.
2024-07-31 15:20:56 +02:00
20bbb411c2
fix(xtts): update streaming for transformers>=4.42.0 ( #59 )
...
* Fix Stream Generator on MacOS
* Make it work on mps
* Implement custom tensor.isin
* Fix for latest TF
* Comment out hack for now
* Remove unused code
* build: increase minimum transformers version
* style: fix
---------
Co-authored-by: Enno Hermann <Eginhard@users.noreply.github.com>
2024-07-25 16:24:10 +02:00
20583a496e
Merge pull request #57 from idiap/xtts-vocab
...
fix(xtts): load tokenizer file based on config as last resort
2024-07-25 13:26:28 +01:00
de35920317
Merge pull request #50 from idiap/umap
...
build: move umap-learn into optional notebook dependencies
2024-07-25 13:26:09 +01:00
9192ef1aa6
fix(xtts): load tokenizer file based on config as last resort
2024-07-05 13:52:01 +02:00
6ea3b75b84
Update xtts.py ( #53 )
...
docs(xtts): fix typo in example
2024-07-02 13:43:52 +02:00
2d06aeb79b
chore: remove unused TTS.utils.io module
...
All uses of these methods were replaced with the equivalents from coqui-tts-trainer
2024-06-29 15:07:10 +02:00
e869b9b658
refactor: use load_checkpoint from trainer
2024-06-29 15:07:10 +02:00
da82d55329
refactor: use load_fsspec from trainer
...
Made automatically with:
rg "from TTS.utils.io import load_fsspec" --files-with-matches | xargs sed -i 's/from TTS.utils.io import load_fsspec/from trainer.io import load_fsspec/g'
2024-06-29 15:07:10 +02:00
0fb26f97df
refactor: use get_user_data_dir from trainer
2024-06-29 15:07:10 +02:00
28296c6458
refactor: use get_git_branch from trainer
2024-06-29 15:07:10 +02:00
59ef28d708
build: move umap-learn into optional notebook dependencies
...
Except for notebooks, it's only used to show embedding plots during speaker
encoder training, in which case a warning is now shown to install it.
2024-06-26 23:53:17 +02:00
4bd3df2607
refactor: remove duplicate get_padding
2024-06-26 11:54:36 +02:00
c30fb0f56b
chore: remove duplicate init_weights
2024-06-26 11:46:37 +02:00
c5241d71ab
chore: address pytorch deprecations
...
torch.range(a, b) == torch.arange(a, b+1)
meshgrid indexing: https://github.com/pytorch/pytorch/issues/50276
checkpoint use_reentrant:
https://dev-discuss.pytorch.org/t/bc-breaking-update-to-torch-utils-checkpoint-not-passing-in-use-reentrant-flag-will-raise-an-error/1745
optimizer.step() before scheduler.step():
https://pytorch.org/docs/stable/optim.html#how-to-adjust-learning-rate
2024-06-26 11:38:25 +02:00
a755328e49
refactor(freevc): remove duplicate sequence_mask
2024-06-26 10:17:04 +02:00
f8df19a10c
refactor: remove duplicate convert_pad_shape
2024-06-26 10:17:04 +02:00
cd7b6daf46
fix: clarify types, fix missing functions
2024-06-26 10:17:04 +02:00
d65bcf65bb
chore(freevc): remove duplicate DDSConv and ElementwiseAffine
...
Already exist as:
TTS.tts.layers.vits.stochastic_duration_predictor.DilatedDepthSeparableConv
TTS.tts.layers.vits.stochastic_duration_predictor.ElementwiseAffine
2024-06-26 10:17:04 +02:00
9f80e043e4
refactor(freevc): use existing layernorm
2024-06-26 10:17:04 +02:00
4d9e18ea7d
chore(stream_generator): address lint issues
2024-06-17 09:52:35 +02:00
2a281237d7
refactor(stream_generator): update code for transformers>=4.41.1
...
In line with
eed9ed6798/src/transformers/generation/utils.py
2024-06-17 09:52:35 +02:00
4b6da4e7ba
refactor(stream_generator): update special tokens for transformers>=4.41.1
...
Fixes #31 . The handling of special tokens in `transformers` was changed in
https://github.com/huggingface/transformers/pull/30624 and
https://github.com/huggingface/transformers/pull/30746 . This updates the XTTS
streaming code accordingly.
2024-06-17 09:52:35 +02:00
81ac7abd58
Merge pull request #47 from idiap/numpy2
...
build: add numpy2 support
2024-06-17 08:48:18 +01:00
bd9b21d946
Merge pull request #44 from idiap/phoneme-cleaners
...
Add multilingual phoneme cleaner
2024-06-17 08:47:15 +01:00
4bc0e75a08
build: add numpy2 support
...
Identified necessary code changes with the NPY201 ruff rule. Gruut is the only
dependency that doesn't support numpy2 yet.
NB: At build time numpy>=2.0.0 should be required to be able to build wheels
compatible with both numpy1+2:
https://numpy.org/devdocs/dev/depending_on_numpy.html#numpy-2-abi-handling
2024-06-16 22:10:33 +02:00
3a20f4725f
fix(freevc): use the specified device for pretrained speaker encoder ( #45 )
...
Fixes coqui-ai#3787
2024-06-16 21:24:03 +02:00
9cfcc0a0f5
chore(cleaners): add type hints
2024-06-14 15:20:04 +02:00
e5c208d254
feat(cleaners): add multilingual phoneme cleaner
...
This doesn't convert numbers into English words.
2024-06-14 15:06:03 +02:00
03de4b889e
docs: fix readthedocs links
...
[ci skip]
2024-06-13 22:48:34 +02:00
29e91f2e77
fix(utils.generic_utils): correctly call now()
2024-05-31 08:39:32 +02:00
77722cb0dd
fix(bin.synthesize): correctly handle boolean arguments
...
Previously, e.g. `--use_cuda false` would actually set use_cuda=True:
https://github.com/coqui-ai/TTS/discussions/3762
2024-05-31 08:39:32 +02:00
a682fa8d56
Merge pull request #33 from idiap/versions
...
Fix XTTS streaming
2024-05-29 14:16:36 +01:00
07cbcf825c
fix(espeak_wrapper): read phonemize() input from file
...
Avoids utf8 encoding issues on Windows when passing the text directly.
Fixes https://github.com/coqui-ai/TTS/discussions/3761
2024-05-29 10:10:05 +02:00
Enno Hermann
Enno Hermann
Shavit
Johnny Street
Daniel Walmsley
Abraham Mathews
ChristianRomberg