Wonderful idea to do this, I had a dabble with the same library a while back but didn't have much success. There was a lot of hallucination and false detections due to weak signals.
My approach was to feed a bunch of audio recordings to the model rather than use the mic. Would this be an option you could utilise i.e. give users the option to either use an audio device (virtual audio cable perhaps?) or some wav/mp3 recordings?
Wonderful idea to do this, I had a dabble with the same library a while back but didn't have much success. There was a lot of hallucination and false detections due to weak signals.
My approach was to feed a bunch of audio recordings to the model rather than use the mic. Would this be an option you could utilise i.e. give users the option to either use an audio device (virtual audio cable perhaps?) or some wav/mp3 recordings?