BACK_TO_FEEDAICRIER_2
OmniVoice drops 600+ language voice-cloning TTS
OPEN_SOURCE ↗
REDDIT · REDDIT// 9d agoMODEL RELEASE

OmniVoice drops 600+ language voice-cloning TTS

OmniVoice is a zero-shot multilingual TTS model from k2-fsa that reports support for 646 languages on Hugging Face, plus voice cloning, voice design, and fast inference. It targets developers who want a local, open-source speech stack instead of a hosted TTS API.

// ANALYSIS

OmniVoice is a serious open-source speech release, but the licensing story looks less clean than the Apache-2.0 badge suggests if the tokenizer dependency really carries Boson AI commercial terms. The 646-language coverage is the standout claim and puts it in a very small class of multilingual TTS systems. Voice design is the most useful product feature here because attribute-based control over accent, age, pitch, whisper, and related traits makes it easier to build consistent voice experiences. The reported RTF of 0.025 is the kind of speed number that matters for local apps, batch generation, and interactive voice agents. If the tokenizer dependency is part of the shipped stack, teams should verify the downstream commercial terms before treating this as fully permissive open source.

// TAGS
speechaudio-genopen-sourceinferenceomnivoice

DISCOVERED

9d ago

2026-04-02

PUBLISHED

10d ago

2026-04-02

RELEVANCE

8/ 10

AUTHOR

HelpfulHand3