Feedback

Followed the instructions to increase the range of the INMP441 connected to my ESP32-S3. Put auto gain from 5 to 31, volume multiplier to 4 then 8 then 16 then 64, increased noise suppression from 1 to 4. Nothing seems to have any effect. I've enabled debugging so I can listen to the recordings when they're sent to HA for processing, and they all sound the same to my ears. When I change the settings, ESPHome recompiles and re-uploads each time, but maybe some cache needs to be cleared for those variables to change?

Also, there is a consistent pop at the beginning of each recording, which may affect the auto-gain, when I normalize it in Audacity, it looks like the following.

The ESP32 is fast enough to stream the audio after the wakeword detection with VERY little delay, and to get rid of the pop, 300ms could be trimmed from the beginning.

If I exclude the pop, and then normalize the speech portion, the quality is good, I think plenty good enough for Vosk to do recognition. Close up, Vosk performs very well.

Any ideas why my mic is so quiet and the parameters don't seem to affect it?

URL

https://www.home-assistant.io/voice_control/troubleshooting/

Version

2024.10.4

Additional information

No response