Building confidence in the delta metric for noise anomalies on Orcasound hydrophones

[scottveirs]

One method of checking the utility and quality of the ambient sound noise metrics is to compare them with more conventional visualizations and analyses of the Orcasound audio data. Let's take notes and discuss how well it is working, how useful it is, and the code base could be improved.

Here is the latest PNG of recent delta metric time series for Port Townsend:

There are about 8 prominent peaks in the relative broadband noise level time series.

And here is the corresponding spectrogram:

The spectrogram also suggests about 8 vessels passed by emitting significant noise. The 2nd peak may have a boat passage near it's start. And the 3rd peak is a double-peak, suggesting two vessels may have passed around the same time.

For a rough comparison, I used my ts2mp3.sh script to pull the latest ~5 hours (5:30-10:30 Pacific) from the Port Townsend S3 bucket and loaded the resulting mp3 file into Audacity. There don't seem to be an obvious data gaps, e.g. silent periods, during this period:

Despite the different look due to logarithmic vs linear display of the frequencies, there also appear to be about 8 peaks in the spectrogram. The raw waveform is much noisier than the delta metrics averaged trends.

Also noticing in the Github Action spectrogram that the logarithmic display hints at a tidal signal in the distribution of lower frequency ambient sound levels. The tidal height at Port Townsend peaked around 3 am and an extreme low tide (the lowest of the year) occurred 10:24 this morning: