From the documentation, "The VAD block monitors the signal from a microphone channel for voice-like patterns and on detection of a matching pattern that triggers an interrupt. The VAD monitors for both an onset of voice-activity as well as the end of voice-activity. Both events can be mapped to interrupts."
During my testing with the ADCx120EVM-PDK I am noticing that the VAD, when in auto-mode, starts recording for any loud noises such as knocking on the table near the device, snapping, or even clapping. Is this the expected behavior? I'm hoping that the VAD can be more sensitive to speech only detection. Can you please comment on the following scenario?
Here is a host recording of the ASI output when introducing a knock on the table near the development board. The audio playback is as such.
Here is a logic capture of the SDOUT pin and the 2ms pulsed interrupt on GPIO1. The analog mic on ADC CH1 starts recording after the first interrupt pulse and stops when the VAD turns off after the next pulse. This leads me to assume that recordings would be captured for any number of mechanical noises along with speech.
I did not expect this behavior for a Voice Activity Detector. The application note makes mention of some tuning registers that may reduce false positives, but they aren't writable using PurePath Console3 register map.
Attached is a register dump that was used during the above scenario. Are there actions I can take to improve speech detection?
