Windows Vista Home Basic
Audacity 2.0.5
.exe installer used
Hello everyone,
I work in a research lab and we record mouse ultrasonic vocalizations using a specialized microphone. Since we cannot know what section of our recording includes just noise (and no specific calls), we cannot choose a noise profile that is guaranteed not to have specific calls in the time window. Is there an alternate way we can find a way to reduce noise in our recording? I have attached a file showing the high amount of noise we experienced for several of the recordings we have. Thank you!
ultrasonic vocalizations using a specialized microphone.
You, unfortunately, fall into the Forensics camp. Make trash into a desirable performance. We can’t do forensics.
People who use Noise Removal under perfect circumstances frequently post that their main show has become damaged or unusable. I would imagine trying to use the tool blind would result in even less luck. Even worse, the tools within Noise Removal designed to prevent audible damage are designed to do it with human speech, not effects sounds.
Wearing my Hardware Hat for a minute, You’re using a special ultrasonic microphone designed for this job…but is it plugged into a sensitive ultrasonic preamplifier and a digitizer up to the job? It’s not unusual for people to try and record out-of-human-hearing work in both directions using regular, ordinary electronics. It’s rarely successful. See: “What’s that rumble I hear coming from next door?”
Small animals that use ultrasonic sound aren’t bashful about it. Given a reasonable microphone, you can hear a bat coming from across the road. If your’e close-miking a mouse, I can well imagine there’s something wrong with the electronics.
Do you have a recording standard? A known, working bat? I expect you do, or you would not be able to publish the work.
I have a request in for a way to generate a Noise Removal profile by averaging out a long performance and assume any signal present for a long time is noise. This seems perfect for your job. It’s a Feature Request along with hundreds of others.
Since we cannot know what section of our recording includes just noise (and no specific calls), we cannot choose a noise profile that is guaranteed not to have specific calls in the time window.
If the signal-to-noise ratio is so bad that you can’t tell when there’s signal+noise or just noise alone, noise reduction isn’t going to help. I assume you’ve tried slowing-down the playback to bring the ultrasonic sounds down’ to where you can hear it?
Noise reduction works best when you have a slight-constant background noise. And, I have no idea if Audacity’s noise reduction works in the ultrasonic range.
You might start with a high-pass filter to filter-out anything below the expected mouse frequency range.
+1 for slowing down the track so you can hear Micky , then you can apply noise reduction tools designed to process audio within human hearing range. You could slow it down quickly by changing the playback sample-rate to something much slower.
I would not only slow the audio down but also scale the frequency part such that it lies similar to a human’s spectrum.
Can you tell us more about the vocalization of mice?
What is the nature of the ultra sonic sounds, do they coincide with audible utterances and form quasi-formants?
Are they for communication, orientation or what?
It is possible that FM-demodulation could reveal more “language” keys and be a true eye opener.
There are several techniques that could improve the result.
noise removal with threshold (e.g. 15 % rule)
Linear prediction coding, where vocal-track information is emphased and white noise diminished.
peak tracking; the xx highest peaks in the target spectrum are tracked and re-synthesized.
Thank you for the responses. It was only this batch of recordings that had high levels of noise, and I figured out the problem. The thread connection btw the microphone module and capsule had come loose. It is working fine now. We have other previous cases and new cases where the recordings worked and we can clearly see a signal (with low noise). I was just curious about anything that can help with this specific set of recordings.
If the signal-to-noise ratio is so bad that you can’t tell when there’s signal+noise or just noise alone, noise reduction isn’t going to help.
I’m assuming this answers that question. If I cannot see any signal amidst the noise, I’m assuming I wouldn’t be able to see it no matter how much noise I can remove?
It’s common advice for podcast producers to make a “Room Tone” segment in a recording. Stop what you’re doing for five seconds and hold your breath while recording. That then becomes the Noise Removal Profile. Literally, the Tones that the Room (and microphone system) is making without you.
Removal will then try to subtract that from your show. If you do a good job, you can create a significant reduction in background noise and it’s startling how much better we can make your show with relatively modest processing.
So if you can make the mouse shut up for five seconds, you got it licked.
There’s almost no chance of success without a good profile. Audacity does not have “understanding” and doesn’t “know” what undesirable sound is. Everything to Audacity is blue waves on the timeline and sampling numbers.
The hearing iss sometimes more sensitive then other senses.
generate pink noise with amplitude 0.25
generate a a Dtmf sequence.
amplify the latter by -32.
You should still be able to hear the tones, although they are 10 times quieter than the noise.
It is all about integration time, for the ears and for any detecting tool as well. The longer the tone lasts, the easier it will be.
It’s also helpful to know the bandwidth of the useful signal.
For the above example, we could filter out frequencies above 700 and below 1600 Hz.
