Radar ·
Phonon 31M: open TTS model in Swift with CoreML on Apple's Neural Engine
A Swift port of the open Phonon 31M text-to-speech model, running CoreML on Apple's Neural Engine, with 103 voices and a text normalizer ported from Gradium. The author reports about 20 ms to first audio and about 20x real time on an M5 Pro.
According to the publication, the project is a Swift port of the open Phonon 31M text-to-speech model, run through CoreML on the Neural Engine of Apple devices. The package includes 103 voices and a text normalizer ported from Gradium. The author claims about 20 ms to first audio and about 20x real time on an M5 Pro chip. These figures are the author's claims, not independent measurements verified by this bulletin.
For engineers on Apple platforms, the material offers an open TTS option that runs on the device itself, with code and model linked from the original source. To check it, open the original link, review the license for the code and model, and reproduce the latency tests on your own hardware before any use.
Original source: Phonon 31M: modelo TTS aberto rodando na Apple Neural Engine em Swift