Show HN: Phonon TTS on WebGPU, 44MB, 115x real time, with live word timings

Category: ai-ml

Tags: webgpu, text-to-speech, browser, webassembly, audio, machine-learning

Score: 8.3/10 (Innovation: 8, Technical: 9, Documentation: 9, Utility: 7)

Phonon is a WebGPU-accelerated text-to-speech runtime for the browser, based on an Apache-2.0 model from Gradium. It achieves 115x real-time speed on an M3 Max while providing per-word timings, and it falls back to a WebAssembly CPU tier. The project includes rigorous benchmarks against alternatives, detailed documentation, and a compact 44 MB model download.

Target audience: browser developers, frontend engineers, AI/ML practitioners, accessibility developers

Repository: https://github.com/blixt/phonon · TypeScript · Apache-2.0 · 22 stars

View on Hacker News