Clarification Needed: Achieving True Real-time "See-While-Hear" Experience

[leonsama]

Hi team! First, I want to commend you on this impressive work.

I noticed a potential discrepancy in the implementation that I'd like to clarify. The README states that "Stream-Omni can offer users a seamless 'see-while-hear' experience." However, the current /token-to-speech API implementation appears to be offline—users only receive audio after the entire token generation completes, rather than during streaming.

To better align with the documented "see-while-hear" capability, could you either:

1. Provide a demo of synchronous audio stream generation (Like LLaMa-Omni2), or
2. Share guidance on implementing true real-time audio streaming?

This would help clarify the expected user experience. Thanks for your time and consideration!

[Rc-H]

Will it provide real-time interaction in a form similar to websocket or webrtc?