v0.27.4

[eginhard]

What's Changed

⚠️ Coqui doesn't install Pytorch dependencies (torch, torchaudio, torchcodec) by default anymore, making it easier to install the right version for your system yourself. See the updated installation instructions.

Features

• Support Python 3.14 by @eginhard in Support Python 3.14, make Gruut dependency optional #543
• Support PyTorch 2.10 by @eginhard in Support Pytorch 2.10 #548
  ⚠️ The following models only work with PyTorch<=2.9:
  • tts_models/fr/mai/tacotron2-DDC
  • tts_models/en/sam/tacotron-DDC
  • tts_models/nl/mai/tacotron2-DDC
  • tts_models/en/ek1/tacotron2
  • tts_models/es/mai/tacotron2-DDC
  • tts_models/zh-CN/baker/tacotron2-DDC-GST
  • tts_models/de/thorsten/tacotron2-DCA

Fixes

• ⚠️ Do not install PyTorch by default anymore by @eginhard in Remove encodec dependency #540
• Fix TTS/bin/compute_attention_masks.py by @eginhard in Fix compute_attention_masks.py, refactor tests #528
• Make notebooks runnable as is with included test data by @eginhard in Make notebooks runnable as is and include them in tests #539
• Remove encodec dependency by @eginhard in Remove encodec dependency #540
• Make gruut dependency optional by @eginhard in Support Python 3.14, make Gruut dependency optional #543

Full Changelog: v0.27.3...v0.27.4

---

This discussion was created from the release v0.27.4.

[ROBERT-MCDOWELL]

good thing it does not install pytorch, but what about torchaudio? is it the same?

[eginhard]

Yes, same for torchaudio and torchcodec. I clarified the message above, thanks!

[jromal]

I was using this fork without problems, using it in several programs that I run with "uv run script.py".
Since today it does not work anymore. Not even with version 0.27.3.

I am starting to become disappointed.

[eginhard]

Could you add a bit more information so I can help resolve the issue? What command are you running and what is the exact error message you see?

[jromal]

Thanks for asking.

There must be a temporary inaccuracy of "uv" or "pypi" to find the right dependencies. I did thousands tests and changes.
And at the end all worked without much changes. Maybe it was only the need to "transformers" as a dependency.

No idea. Now at least some of the programs work.

A list with stable versions of dependencies that work, would not be a bad idea to share.
Also some documentation on how to load a local model would help. I could not find it on the documentation. And my main usage is cloning voices (with the old "xtts-finetune-webui") and then using them in programs that generate lots of sentences for long audio files, using xtts manual inference. But that's another topic.

Thanks again. I am fine for the moment.

[eginhard]

The minimum supported versions of all dependencies are listed here:

coqui-ai-TTS/pyproject.toml

Line 59 in 36c6ff8

dependencies = [

Every change is tested for that and the most recent version of each dependency, so generally things should "just" work, but edge cases can always happen, e.g. when a new PyTorch or transformers release contains backwards-incompatible changes. In case of any issues in the future, just share any error messages here, without that it's impossible to give concrete help.

To load a local model:

from TTS.api import TTS

tts = TTS(model_path="path/to/model.pth", config_path="path/to/config.json")

# or for XTTS (the next version will support the other way as well)
tts = TTS(model_path="path/to/", config_path="path/to/config.json")