No Acoustic Echo Cancellation

[Galaxywalk]

It seems that the current pipeline is almost linear and no acoustic echo cancellation is implemented. It means that when the speaker speaks the responses from LLM, the microphone may record this audio incorrectly?

[eustlb]

Hey @Galaxywalk,
The pipeline is implemented such that as soon as an end of speech is detected by the VAD, it stops listening to the microphone until the spoken LLM answer has been fully played.
Echo cancellation is the one big feature we would love to implement because it would enable interruptions.

[Galaxywalk]

I'm currently working on acoustic echo cancellation for a language-instructed robot, which I believe is crucial. Could you share your schedule for this feature? Also, packaging the current pipeline into a ROS package would significantly enhance its usability and maintainability for ROS users. Do you have any plans for this?

I've already made some progress in this area and would be eager to contribute. What are your thoughts on external contributions, and would it be possible for an external PR about it?