method to detect frequency sweep in audio

Question

I have search a bit to try to find a similar questions, but I have not been able to find any leads.

There are techniques to programmatically detect steady tones in audio, such as the Goertzel algorithm. I have been trying to find a way to detect a frequency sweep of some kind. My specific application is to monitor the audio from a security camera, and detect, say, when a door prop or car alarm is triggered. If it matters, one can assume that the sound will repeat many times, and that the frequency is a smooth sweep, and not discreet tones.

I don't want to base it strictly off of amplitude, since the frequency response is sometimes very non-linear on these cheap cameras, and other noises can be a source of false alarms. Also, the alarm source may be distant.

Answer 1

If you know the signal of interest in advance, you can pre-record it and constantly calculate correlation between recorded sample and incoming signal.

• Record a sample sound from a channel to be analyzed (you can record a number of periods)
• Reverse it
• Use reversed signal as an impulse response for FIR filter (this also is known as "matched filter") - filter will output a peak when signal of interest passes through it.

If you are interested in general varying tone detection (much more difficult problem), look at the chirplet transform - the sort of wavelet transform which uses varying frequency wavelets (chirplets) as its basis. People used it to count wolf howls, for example.