supertone-inc/supertonic

Lightning-Fast, On-Device, Multilingual TTS — running natively via ONNX.

What it solves

Supertonic is a high-performance, on-device text-to-speech (TTS) system that eliminates the need for cloud APIs, reducing latency and improving privacy. It provides a lightweight alternative to massive TTS models, allowing studio-grade audio synthesis to run locally on resource-constrained hardware like mobile devices, browsers, and Raspberry Pi.

How it works

Powered by ONNX Runtime, Supertonic uses a compact 99M-parameter open-weight model to perform inference entirely on the device. It supports 31 languages and can operate in a language-agnostic mode (lang="na") when the input language is unknown. The system outputs 44.1kHz 16-bit WAV audio and includes a set of expression tags (e.g., <laugh>, <breath>) to add human-like nuance to speech without requiring complex prompt engineering.

Who it’s for

Developers building local-first applications, such as browser extensions, e-readers, and edge-device software, who need fast, private, and high-quality speech synthesis without relying on a GPU or network connection.

Highlights

  • Edge-Ready: Runs locally on desktop, mobile, browsers (via WebGPU/WASM), and Raspberry Pi.
  • Multilingual: Supports 31 languages with an optional language-agnostic processing mode.
  • Compact Model: 99M-parameter footprint is significantly smaller than typical 0.7B-2B class TTS systems.
  • Broad SDK Support: Ready-to-use examples for Python, Node.js, Java, C++, C#, Go, Swift, iOS, Rust, and Flutter.
  • Natural Text Handling: Specialized in accurately reading financial expressions, phone numbers, and technical units.
  • Expression Tags: Built-in tags for non-verbal sounds like sighs and laughs to enhance realism.

Related

  • Project
    OpenMOSS/MOSS-TTS-Nano

    A multilingual tiny speech generation model with 0.1B parameters designed for real-time, CPU-based inference and high-fidelity audio output.

  • Project
    kyutai-labs/pocket-tts

    A lightweight, CPU-efficient text-to-speech application that enables local voice cloning and multi-language audio generation without requiring a GPU.

  • Project
    resemble-ai/chatterbox

    A family of open-source text-to-speech models providing low-latency voice cloning and multilingual speech generation across various model sizes for GPU and CPU deployment.

  • Project
    KittenML/KittenTTS

    A lightweight, ONNX-based text-to-speech library that provides high-quality voice synthesis on CPU with models as small as 25MB.

  • Project
    neuphonic/neutts

    A collection of open-source, on-device text-to-speech models that enable natural-sounding speech and instant voice cloning locally on mobile and embedded devices.