We have added a feature to provide an overlay on the screen, displaying motion/face detection, microphone audio level and wake word detection value to aide with voice issues - be it not hearing wake word, too many false positives, poor interpretation of commands or just pure geeky interest!

This feature can be enabled/disabled using the On screen diagnostics switch entity in the VACA integration device.

Please read the below to understand what this is showing you.

Motion detection

Motion detection is only available when the device has a supported front facing camera. This overlay will show when motion detection is enabled and turn green when detected. If the mode is set to Face detection, it will stay green as long as it detects a face.

Microphone audio level

This is the level of audio being processed by the wake word engine (while waiting for wake word) or being sent to the Speach To Text (STT) engine when the wake word has been detected.

You will notice on some devices that the audio level seems very low (in the 0-2 level) and on others much higher (in the 20-30 level) when listening for the wake word in the Detecting mode. This is ok (is a feature of the device hardware/OEM Android config) and the wake word detection engine will work just as well.

Note

The gain setting makes no difference to the audio level during the wake word detection phase.

Once the wake word is detected, VACA switches into Streaming mode to stream audio over the network to HA and you will see a significant jump in the audio level (in the 20 to 50 range). This is where the auto gain function has kicked in to boost the audio level (to try and be consistent no matter the device) and improve the command interpretation by your chosen STT engine. The mic gain setting is in effect here. You should expect this level to be in the high 20 to 30 range with gain at 0. Dropping to 10 at -10 and increasing to 50 at +10.

So, why not boost the audio level for wake word detection? Well, very good question. And the answer is, we tried that and it actually made it worse in real world testing, so we removed it.

Wake word Detection

This is the wake word engine's level of confidence that the audio heard was the wake word when in wake word listening mode. In the settings, there is a wake word threshold setting, which directly relates to this detection level number. Ie, when this detection level number is equal to or greater than the threshold level, it will class that as a wake word detection.

A setting of 6 on the threshold setting should be a good place to start to find the right balance for good detection without many false detections in noisier environments.

Some wake words are better at detecting/less sensitive to accents than others and you may need to adjust this threshold setting if you change the wake word you wish to use. And this diagnostic overlay should help you get that setting right for you.