Skip to content

piper.md

Latest commit

 

History

History
56 lines (42 loc) · 1.74 KB

piper.md

File metadata and controls

56 lines (42 loc) · 1.74 KB
Raw

Piper Engine

PiperEngine shells out to the Piper executable and streams raw PCM from a Piper ONNX voice model. It is a practical local deployment path when you can ship a Piper binary and model files.

Install

Install RealtimeTTS with the Piper extra, then set up Piper separately. The Python extra only covers RealtimeTTS-side dependencies; it cannot install the Piper executable or voice model files.

pip install "realtimetts[piper]"

Download or build a Piper executable and a Piper voice model. You can pass the executable path to PiperEngine or set PIPER_PATH.

$env:PIPER_PATH = "D:\path\to\piper.exe"

Minimal Use

from RealtimeTTS import TextToAudioStream, PiperEngine, PiperVoice


if __name__ == "__main__":
    voice = PiperVoice(
        model_file="voices/en_US-lessac-medium.onnx",
        config_file="voices/en_US-lessac-medium.onnx.json",
    )
    engine = PiperEngine(voice=voice)
    stream = TextToAudioStream(engine)
    stream.feed("Hello from Piper.")
    stream.play()

Source Notes

  • PiperVoice(model_file, config_file=None) derives config_file from model_file + ".json" when possible.
  • PiperEngine(piper_path=None, voice=None, debug=False) uses PIPER_PATH and then piper.exe when piper_path is omitted.
  • The engine runs Piper with --output-raw and feeds text through stdin.
  • Sample rate is read from the voice JSON config; fallback is 16000 Hz.
  • get_voices() returns an empty list because Piper voices are local files, not discoverable through the wrapper.

Troubleshooting

  • Piper executable not found: set PIPER_PATH or pass piper_path.
  • If speech plays at the wrong speed, check that the matching .json config is passed with the ONNX model.