🐍 Python API

Install with:

pip install piper-tts

Download a voice, for example:

python3 -m piper.download_voices en_US-lessac-medium

Use PiperVoice.synthesize_wav:

import wave
from piper import PiperVoice

voice = PiperVoice.load("/path/to/en_US-lessac-medium.onnx")
with wave.open("test.wav", "wb") as wav_file:
    voice.synthesize_wav("Welcome to the world of speech synthesis!", wav_file)

Adjust synthesis:

syn_config = SynthesisConfig(
    volume=0.5,  # half as loud
    length_scale=2.0,  # twice as slow
    noise_scale=1.0,  # more audio variation
    noise_w_scale=1.0,  # more speaking variation
    normalize_audio=False, # use raw audio from voice
)

voice.synthesize_wav(..., syn_config=syn_config)

To use CUDA for GPU acceleration:

voice = PiperVoice.load(..., use_cuda=True)

This requires the onnxruntime-gpu package to be installed.

For streaming, use PiperVoice.synthesize:

for chunk in voice.synthesize("..."):
    set_audio_format(chunk.sample_rate, chunk.sample_width, chunk.sample_channels)
    write_raw_data(chunk.audio_int16_bytes)