“Normally I do not speak much when I’m sleeping, really only when I’m awake.” 
“How can I share my recordings - by email? Can you explain the reference to logic you made earlier please?”
My colleague just tried your recordings and indeed the old neural network produced a lot of false positives, but the new network seems to work well and does not miss-classify them … big thanks for your help on this …
Maybe one reason is the noise threshold setting which does not work? I increased the threshold every night by 10%, my current setting is 50%, but it looks like the setting has no effect. I can’t exclude low noise events from recording.
Please make the noise threshold settings work and show the noise level in noise recording diagrams. Currently noise diagrams look similar for very low volume recording and for high level volume recording.
This will not solve all issues with wrong talk detections because they also occur on high volume recordings. But maybe you should pre filter noise data for talk detection a little bit more by volume, if low volume talk is one reason of false detection.
Hello, many thanks, I happy to hear the classifier improved, but honestly every device is different does different audio pre-processing and also the HW is different so this always will be device specific although we do maximum we can to reduce the device specific effects on the classifier…
Regarding the threshold it should work, although again volume level are device specific and there is no way to actually calculate decibels of the received sound to make it work the same on all devices… but you can easily test this… if you enter the sleep tracking screen you see a volume bar right of the red dot of the recording indicator…
There are two lines on the progress bar, one shows actual voluem level and the other shows the threshold, so you can see how that works for you… just make some noise and see if it gets over the threshold or not…