Supertonic

Supertonic is designed as a highly efficient TTS stack that trades bulky architectures for a compact 66M parameter model optimized around ONNX Runtime. All synthesis happens on device, so there are no cloud calls or server round trips, which gives near zero latency and strong privacy guarantees. The system handles natural text directly, including numbers, dates, currency, abbreviations, and other tricky patterns without heavy pre processing, and it reaches speeds reported at up to roughly 167 times faster than real time on modern consumer chips. Supertonic ships with ready made inference examples across Python, Node.js, web (WebGPU or WASM), Java, C or C++, C sharp, Go, Swift, iOS, and Rust, making it straightforward to embed in desktop apps, mobile clients, browsers, or edge devices. Configuration options let developers tune batch size, decoding settings, and throughput to match their hardware and latency budgets, turning Supertonic into a practical base for assistants, read aloud features, accessibility tools, and fully offline voice experiences.