Closed collvey closed 2 years ago
Actually the compute_f0() is still being used in https://github.com/coqui-ai/TTS/blob/f237e4ccd9f2fd2b0cb5e136dfe0e20cc32bf898/TTS/tts/datasets/dataset.py#L716 . Will close to find other solutions.
Reopen with updated description. Proposes to remove requirement of pyworld
for TTS pip installation.
This is a problem on other platforms as well. In Fedora and Debian the "pip install -e .[all,dev,notebooks]" method fails.
This was fixed in Fedora 33 by running "sudo dnf install python3-devel". On Debian 11, even though python3-dev was installed the pyworld module failed to build.
@WeberJulian is investigating an alternative to pyworld. Unfortunately, until we find one, this issue remains to stay.
To get rid of this we can switch to pyin. It is slower but it works :).
Would you deal with that @Edresson?
This issue was solved with PR #1946 where we replaced the Pyworld with Librosa Pyin.
Describe the bug
Environment: OS Name Microsoft Windows 10 Pro Version 10.0.19044 Build 19044 Python 3.9.13
Running
pip install TTS
in Windows triggers a known build error for one of its dependencies:pyworld
https://github.com/JeremyCCHsu/Python-Wrapper-for-World-Vocoder/issues/73The pyworld build error is fixable by installing correct MSVC toolset (Microsoft Visual C++ 14.0 or greater is required.) in https://visualstudio.microsoft.com/visual-cpp-build-tools/ but will lead to another link error possibly related to Cython version (similar issue: https://github.com/wxWidgets/Phoenix/pull/2017)
A cleaner way is to remove dependency on
pyworld
for TTS pip installation. The pip installation option is for people who are only interested in synthesizing speech with the released 🐸TTS models, not for training the model. Correct me if wrong, there is no need to depend onpyworld
which is, as I understand, only for training the model.To Reproduce
Expected behavior
No response
Logs
No response
Environment
Additional context
No response