Speech-token inversion clones a voice from a short leaked representation
A three-second Moshi leak produced 0.78 average speaker similarity and a 3.80% target-text word error rate.
Stories per day · 30 days
3 RSS
4 stories, newest first · one card per story
Stories tagged #moshi
HiPLEX: Hierarchical Policy Factorization for Full Duplex Speech Language Models
HiPLEX factorizes full-duplex speech policies so reinforcement learning can improve timing and content separately.
Tokens Change, Structure Endures: Spectral Watermarking for Generated Speech
Redwing keeps generated-speech watermarks intact through retokenization using a spectral substitution basis.
SteerDuplex: Steerable Duplex Speech Dialogue Models
SteerDuplex fine-tunes the Moshi full-duplex speech model with two-stage RL and hybrid rewards to steer tone, persona, and pacing, alongside the new SteerBench benchmark.