StyleTTS2 Inferencer

This python package is a wrapper around the very nice StyleTTS2 TTS engine made by Aaron (Yinghao) Li et al. It currently supports converting a specified text into speech which is stored to a specified speech file.

How to install:

• sudo apt-get install espeak-ng git
• git lfs install
• pip install . The model weights will be automatically downloaded during the first run.

Run with:

from styletts2_inferencer import StyleTTS2Inferencer

tts = StyleTTS2Inferencer() # defaults to using cuda/gpu
tts.tts_to_file(
    text="Hello! How are you?", # automatically switches to long form generation for multiline texts.
    filepath="test.wav" # should be an absolute path
)
# Output: 
# RTF = 0.410011
# Written output to file test.wav

See test.wav for an example output.

Improvements to make

• Add parameter for speech expresseviness and number of diffusion steps as shown in the demo