VoxCPM

VoxCPM is a text-to-speech system that turns text into speech without discrete tokenization—that is, without first breaking speech into discrete tokens—using an end-to-end diffusion autoregressive model to generate continuous speech representations. ^[1] VoxCPM2 can synthesize speech in 30 languages and supports voice design, which creates a voice from a natural-language description, and voice cloning from a short reference clip. ^[2a][2b] A supported use case is generating an audio file from text, optionally with a reference audio path to clone a voice or style guidance to adjust emotion, pace, or style. ^[3]