search

orcasound / orcanode-monitor

Web service for monitoring liveness of orcanode audio streaming
MIT License
0 stars 1 forks source link

Intelligibility detection algorithm is flaky #66

Closed dthaler closed 1 month ago

dthaler commented 4 months ago

PR #55 added an algorithm but it generates false Unintelligible signals multiple times a day

scottveirs commented 2 months ago

Does "Intelligibility" mean the audio data as a zero-amplitude signal, i.e. the data files are being created and uploaded but the hydrophone signal is absent in the time series?

If so, a possible short term solution would be to define a "no signal" status as a .ts segment where the standard deviation of the RMS amplitude is zero (not under some threshold).

A longer-term solution would be to wait until shipnoise.net or a https://github.com/orcasound/orca-action-workflow deployed for all nodes is generating a recent time series of the sound pressure levels for short-term means, like by hour, minute, or second. Then use those time-series to detect zero standard deviation periods.

dthaler commented 2 months ago

Point Robinson isn't 0 amplitude, it seems to be a microphone hum with no actual signal but not just silence. That makes it trickier to distinguish between a hum and normal steady state.

dthaler commented 1 month ago

Fixed by PR #137