I am testing the functionality of the VAD on ADC CH3 using the ADCx120EVM-PDK development board. Part of my evaluation is playing several audio database files over a speaker in close proximity to the development board to simulate conversation. For short and isolated speech patterns the VAD speech detection and recording is accurate. For the attached WAV file however, there is a sample dropout at the beginning and middle of the sentence. During these windows speech is not recorded. The first might be attributable to latency in speech detection and/or ADC power-on time which is acceptable if only at the onset of speech detection. It is in the second window where the VAD turns off prematurely and some speech is not recorded.
In this image the top waveform was captured with VAD disabled for the attached WAV file. The bottom is with VAD auto-mode enabled.
This image is a logic output of the SDOUT pin and the GPIO1 interrupt pin identifying when the VAD and consequently ADC CH1 turns on and off. My experience was that the VAD turned off quickly during a brief pause in the speech, yet it remains on for a long time even after speech has ceased. The latter is expected as we want to record until speech is no longer detected. Can anyone explain why the former happens so promptly?
Attached is the speech test file used and a register dump used during this scenario. The only modifications were enabling/disabling VAD from the GUI.
