Commit

Basically, I aim for a single real-time model, but the transcript can be later post-processed with LLM if needed. re. model, see: - https://help.aliyun.com/en/model-studio/real-time-speech-recognition-user-guide?spm=a2c4g.11186623.help-menu-2400256.d_1_2_2_0_1.362dc690INtcxw… - https://github.com/dtinth/vxbeamer/tree/main/packages/vxasrdtinth/vxbeamer:vxasr/ (it has adapter for Qwen audio already) re. directions: - The transcript will be public, i.e. anyone with the link can access. - Contrary to v1, VAD should be done server-side. - Should double as audio recorder so we can re-transcribe if needed. - Built for robustness. - For sender, it should keep a local buffer of audio the moment recording started, while connecting to WS in the background. Should a connection be lost, the client keeps recording and can catch up with the sender. - Can treat audio as 100ms chunk. Each chunk has a timestamp. This way if connection lost and we reconnected, the server can tell which chunk was last received. - Audio compression is desirable but not strictly needed. Please install https://github.com/mattpocock/skillsmattpocock/skills and set up the project.