Detecting faint sounds at certain frequencies?

[brian0918]

I have large 4-8 hour wav files that are mostly outdoor background noise (recorded at night), but there will be occasional short, faint sounds in the 2-10 kHz range. Is there a function in Audacity, or a plugin, that would be able to find these sounds, and jump to them?

I currently have Windows 7 SP1 and Audacity 2.2.2.

[steve]

You may be able to see them in the Spectrogram track view: https://manual.audacityteam.org/man/spe ... _view.html

[kozikowski]

that would be able to find these sounds

Make sure you have protection copies of all this work on separate media before you start messing with filters and effects. You should be able to point to two different storage places as proof of valuable work backup.

I think I would trim one recording down to about a half-hour around one sound you know is there until you get a technique to work. Otherwise you could retire waiting for Audacity to finish testing, not actually searching.

I think I would first do a noise reduction using a non vocal segment of the recording as Profile. Then apply it to the whole recording. That should get rid of microphone and other standard noises. If everything vanishes, or is suppressed, that's not going to work.

Brute force filtering.

Effect > High Pass Filter: 2KHz, 24dB.
Effect > Low Pass Filter: 10KHz, 24dB.

What's left after that should be little haystacks of vocalizations between those two pitches. It's up to you to scan down the blue waves and look for the haystacks. It should be easier because they stick up now.

Past that I got nothing. The human ear is pretty good at pulling quiet sounds out of trash. Trying to get Audacity to do that is a challenge.

I think there are other more magic techniques, but they may only locate the vocalizations, not let you actually listen to them. You still have to write down the time and transfer that to a clean track.

All of these techniques will work better the more accurate you can determine the pitch of the vocalizations. We can't, for example, pull a human voice out of trash. Too many different tones.

Koz

[brian0918]

OK thanks, this is working pretty well.