Audio Signal Processing [ detect/recognize environmental sound ]

Please I am new in signal processing. I want to know the criteria to detect / recognize environmental sound in an audio signal. I have the codes for the extraction techiqiues I want to know already, but I want to know how to use them (Gammatone Cepstral Coefficient and Matching Pursuit ) feature extraction techniques for the detection and recognition of environmental. I need this for my research programme. I will be expecting your prompt response. Thank you

Re: Audio Signal Processing

It should be easy to detect a particular frequency, or possibly a band of frequencies. (Example, sirens, musical instruments, bird songs, voices.) Bandpass filters are able to do this.

Brief loud waveforms should be easy to detect. (Example, hand clap, door slam, drum kick.) A commercial item called 'The Clapper' contains circuitry which is able to detect this type of waveform.

Complex sounds are much more difficult to extract. It helps if you can notice anything distinctive about its waveform, so that you can create a filter or code.

Re: Audio Signal Processing

Thank you, please, how does Bandpass filters detect this? Is there any frequency threshold for this? And how do l obtain my frequency values from MP and GFCC. Thank you

Re: Audio Signal Processing

Filters are applied digitally by FFT methods. It can be for any frequency or band of frequencies, although you must take the sampling rate into account. (Sorry, I don't know further details.)

To examine audio spectrograms, there is a popular free program called Audacity. It displays the entire audible range, in color. From bass frequencies to 20 kHz or higher. You can pick out occurrences of individual notes, from a voice or instrument.

Harmonic overtones are visible. You can tell if a flute is playing, or a violin. It may be possible to write code that is able to pick out these differences.

Audacity contains a digital filter. Its envelope can be tailored to any shape you wish.

Re: Audio Signal Processing

Aruorezi said:

Please I am new in signal processing. I want to know the criteria to detect / recognize environmental sound in an audio signal

Click to expand...

There is no general answer to general question in termos of signal processing, unless you are sure that are using the correct method for an already known sound; You need to know how behaves the environmental sound facing to "the sound of interest", and have to make a lot of experiments with different approaches in different domains, which also do not impede you from using a combination of more than one algorithm. I think the first step for now is to find a periodical pattern on the mixed signal, what essentially involves determining the lowest frequency at which the function will take the waveform samples.

Re: Audio Signal Processing

Thank you, please I don't know if you can show me how to find a periodical pattern on the mixed signal using Gammatone filters and MP. or how to get the frequency from the two techniques to determine an environmental sound e.g clapping sound.

Thank you

Sorry, I have no experience with those tools. We can expect that any method will require knowledge about characteristics of sounds, and what makes one type of sound different from another.

OK, without those tools/techniques, do you know the minimum or maximum frequency value to detect any environmental sound or min/max energy value to detect env. Sound.

Thanks

For birdsongs, crickets, frogs, I imagine the range from 200 to 2000 Hz is right. However this is a wide range. A lot of noises are in this range.

It helps if you know ahead of time what sound you want to detect. Suppose you want to detect geese honking. Their pitch is in a narrow range (less than 1/2 octave). Their honk lasts a fraction of a second. It should be easy to detect them. But you also might pick up similar sounds such as dogs barking, etc.

There are white noise or pink noise sounds such as rainfall, streams, waterfalls. These occupy a wide area in the upper range of audible frequencies. Their sound is not very distinctive. As a result you might also pick up a jetliner flying overhead.

To do audio signal analysis, a spectrogram is a valuable tool. That is what makes Audacity remarkable. You can see the frequency of sound pitches (although the preferences require some experimentation to get the scale of readings you want). No doubt the internet has digitized audio clips of the sounds you want to detect. Audacity can read many audio formats.

As said above, you have to make several experiments by yourself, and to use a certain level of inference to better understand what come up from the signal you have to deal. The process of choosing appropriate algorithms for each case involves a certain degree of freedom, and you must have an open mind to explore all possibilities. This is one of the foundations of data mining, each dataset asks to be treated differently.

ok, according to my project, I could pick any environmental sound so far its not speech and music. that is any other non speech sound will be ok, so I only need frequencies, energy and scale values for environmental sound.

Thank u so much