Issues with voice activity detection

[dgube1]

Experiencing issues with calls to the function vad_process. Application runs fine until function is called resulting in a stack overflow and not sure how trouble shoot due to the source being precompiled binaries. Usage follows the simple example case outlined in the header file but have not had success.

Have any other users had more success in using this library and getting voice activity detection on the esp to work, or can anyone from Espressif provide some insight into usage or when more support will be added for this library and functionality?

[ESP_krzychb]

Hi dgube1,

VAD example is prepared and currently in review for release in esp-adf repository.

[mgsscorr1]

Hello krzychb,

While attempting to run the VAD example, it has come to my conclusion that the "Voice" activity detector fires with any and ever sound that occurs.

Through testing of every single combination of the following settings VAD_MODE_#, VAD_SAMPLE_RATE_HZ, VAD_FRAME_LENGTH_MS
Nothing seems to solve the issue.

Are there any plans to have the VAD up and running properly or more efficiently?

[jason.mao]

Hi mgsscorr1

The VAD algorithm is detect the amplitude of voice relative to noise , if the noise is drastic change, it can be trigger the VAD, e.g. Tap on a table.