Comment by pugio

There is no TTS here. It's a native audio output model which outputs audio tokens directly. (At least, that's how the other real-time models work. Maybe I've misunderstood the Qwen-Omni architecture.)