Coqui TTS

Text-to-speech extension for Oobabooga's text-generation-webui using Coqui TTS.

Installation

Assuming you already have the WebUI set up:

1. Install eSpeak-NG and ensure it is in your PATH
2. Activate the conda environment with the cmd_xxx.bat or using conda activate textgen
3. Enter the text-generation-webui/extensions/ directory and clone this repository

cd text-generation-webui/extensions/
git clone https://github.com/Fire-Input/text-generation-webui-coqui-tts coqui_tts

4. install the requirements

pip install -r extensions/coqui_tts/requirements.txt

Notes

• The coqui_tts extension will automatically download the pretrained model tts_models/en/vctk/vits by default. It is less than 200MB in size, and will be downloaded to \home\USER\.local\share\tts for Linux and C:\Users\USER\AppData\Local\tts for Windows.
• When running oobabooga, the tts package (version TTS==0.17.4) may throw an error about numpy if you are using python < 3.11, try pip install numpy==1.24.4 and pip install numba==0.57.1 to install the most compatible version of numpy and numba for this version. Ignore any error messages about incompatible package versions as the tts package needs to update its requirements.txt to later versions of numpy and numba and restart the WebUI.
• Custom models are not supported yet.
• Everytime you generate a new audio, Coqui will print out a log message to the console. This is normal and unfortunately cannot be disabled.
• Audio files are saved to text-generation-webui/extensions/coqui_tts/outputs/
• A lot of the code is copied from the ElevenLabs extension.
• And some code copied from da3dsoul's fork.
• I do not have a Coqui Studio API key, so I cannot test it. Therefore, it is not supported yet.

Testing Environment

• Windows 11
• Conda Installation with WSL2
• WSL2 Ubuntu 22.04
• Python 3.9.16
• numpy==1.21.6
• Conda 23.3.1
• CUDA 11.7
• WebUI commit: 68dcbc7ebda3f0d9700dde43d0d29324f5c244b1
• eSpeak-NG 1.50