Flashlabs releases the world’s first open-source voice cloning model

The end-to-end speech-to-speech claim is interesting, especially avoiding the ASR→LLM→TTS pipeline, which is where most latency and error compounding happens.