DatSpace
commented
2 years ago The problem was that the configuration for autoreconf had a "patch" for the previous version of autoreconf that was fixed in the latest version. Ubuntu 22 comes with the latest version of the software so there was a mismatch.
In any case, fixed with b05afcdfd8c392ec58eb13cbde19a93c083619db
Libfvad, the webrtc library we were using for VAD seems to be stale for a while and it has problems building on ubuntu 22. Further than that, it seems to be based off an old commit of the webrtc library. A quick search for webrtcvad (The old python version we were using) shows that the fork is around 2016.
https://github.com/wiseman/py-webrtcvad
And : https://github.com/wiseman/py-webrtcvad/issues/69
To be fair we are using the CPP version which even though we had performance improvements, we still had a lot of false positives. https://github.com/dpirch/libfvad
Proposed solutions:
Obviously the microphone plays an important role so as discussed we can look more into that.
Try and maintain our own fork of libfvad OR create a fork directly from google's webrtc (cpp) to use in the project.
Switch back to Python (this time, multi-threaded) using an alternative like https://github.com/snakers4/silero-vad#vad-quality-metrics-methodology