Supertonic — Lightning-Fast On-Device Multilingual TTS via ONNX

Introduction

Supertonic is an on-device text-to-speech engine built for speed and portability. It uses ONNX Runtime for inference, enabling natural-sounding speech synthesis across multiple languages without requiring a network connection or GPU.

What Supertonic Does

• Generates natural speech from text in 12+ languages on-device
• Runs via ONNX Runtime with no cloud dependency required
• Provides native bindings for Rust, Python, Swift, Java, Go, C#, and more
• Supports WebGPU for browser-based inference
• Delivers low-latency synthesis suitable for real-time applications

Architecture Overview

Supertonic packages pre-trained TTS models in ONNX format and runs them through ONNX Runtime on the target platform. A lightweight text processing pipeline handles phonemization and prosody, then feeds tokens to the neural vocoder. Platform-specific bindings wrap the core Rust engine via FFI, keeping the API consistent across languages.

Self-Hosting & Configuration

• Install via pip, npm, cargo, or platform-specific package managers
• Models are bundled or downloaded on first use (typically 50-200 MB per language)
• Configure voice, speed, and pitch through API parameters
• No API keys or cloud accounts needed
• Runs on CPU by default; GPU acceleration available via ONNX Runtime providers

Key Features

• Sub-second latency for short utterances on modern hardware
• Supports Rust, Python, C#, Java, Go, Swift, Ruby, PHP, Elixir, and WebAssembly
• Multilingual support covering major world languages
• Small model footprint suitable for mobile and embedded deployment
• Apache 2.0 licensed with no usage restrictions

Comparison with Similar Tools

• Kokoro — lightweight 82M-parameter TTS; Supertonic focuses on broader language coverage and cross-platform bindings
• Bark — generates speech with music and effects; Supertonic prioritizes speed and on-device deployment
• F5-TTS — flow-matching approach; Supertonic uses ONNX for maximum portability
• Fish Speech — multilingual but Python-focused; Supertonic offers native bindings in 10+ languages
• Piper — fast local TTS; Supertonic provides more language bindings and WebGPU support

FAQ

Q: Does Supertonic require a GPU? A: No. It runs on CPU by default, with optional GPU acceleration through ONNX Runtime execution providers.

Q: What languages are supported for speech synthesis? A: English, Korean, Japanese, Chinese, Spanish, French, German, Italian, Portuguese, Hindi, Thai, and Vietnamese among others.

Q: Can I use it in a mobile app? A: Yes. Native bindings for Swift (iOS) and Java (Android) are provided, along with compact model sizes suitable for mobile.

Q: Is it free for commercial use? A: Yes. Supertonic is released under the Apache 2.0 license.

Sources

• https://github.com/supertone-inc/supertonic
• https://supertonic.ai