supersonic-tts

Stdin-friendly CLI + Python API wrapper for Supertonic — lightning-fast, on-device, multilingual TTS via ONNX Runtime.

No cloud API. No data leaves your machine. 31 languages. 10 voices.

Why supersonic-tts?

Supertonic's official CLI takes text as a positional argument — awkward for piped input, shell scripts, and AI agents. supersonic-tts wraps it with:

• Stdin support — pipe text directly: echo "hello" | supersonic-tts -o out.wav
• Hermes Agent integration — drop-in command-based TTS provider
• Same engine — uses supertonic under the hood (supertonic-3, ONNX)

Install

pip install supersonic-tts

First run auto-downloads the model (~305MB) from HuggingFace.

Usage

CLI

# Read from argument
supersonic-tts "Hello world" -o output.wav

# Read from stdin (piped)
echo "Hello from stdin" | supersonic-tts -o output.wav

# Read from file
cat long_text.txt | supersonic-tts -o output.wav

# Choose voice
supersonic-tts "Crisp and confident" -o output.wav --voice F4

# Multilingual
supersonic-tts "Bonjour le monde" -o french.wav --lang fr --voice F1

# Adjust speed
supersonic-tts "Fast talk" -o fast.wav --speed 1.5

# Higher quality
supersonic-tts "Premium quality" -o premium.wav --steps 10

Python API

from supersonic_tts import SupersonicTTS

tts = SupersonicTTS()
wav, duration = tts.synthesize("Hello world", voice="F4", lang="en")
tts.save("output.wav")

Hermes Agent Integration

Add to ~/.hermes/config.yaml:

tts:
  provider: supersonic-tts
  supersonic-tts:
    type: command
    command: supersonic-tts -o {output_path} --voice {voice} < {input_path}
    voice: F4
    model: supertonic-3
    output_format: wav
    voice_compatible: true
    timeout: 60

Voices

Languages

31 languages (supertonic-3): en, ko, ja, ar, bg, cs, da, de, el, es, et, fi, fr, hi, hr, hu, id, it, lt, lv, nl, pl, pt, ro, ru, sk, sl, sv, tr, uk, vi, na (fallback)

License

MIT (code) — supertonic model uses OpenRAIL-M