Radar ·
Phonon: open 31M-parameter TTS running in the browser via WebGPU
A post reports porting Phonon, an open text-to-speech model with 31M parameters from GradiumAI, to run fully in the browser via WebGPU in half precision, at about 47x real time on an M4 Pro and 7x on a Phone 13 Pro.
According to a post about porting Phonon, an open text-to-speech model with 31M parameters from GradiumAI, the model now runs entirely in the browser through WebGPU, in half precision. The author cites about 47 times real time on an M4 Pro and 7 times on a Phone 13 Pro. These are the author's claims, which this note has not independently verified.
The relevant point is the demonstration of a small voice model running on the client, on common hardware, which may matter for on-device voice features. The source gives no details on audio quality, languages, license or measurement method. To consult the original, search for the post by its title, check the model's license, and reproduce the figures on your own hardware before any production use.
Original source: Phonon: TTS aberto de 31M de parâmetros rodando no navegador via WebGPU