What does an audio file consist of?

[DavidSaunders9]

Hi

My name is David Saunders and I am currently in the middle of a
project looking at being able to compare two audio files and see if
they arethe same song or not. To be able to do this I wish to pick out
the most important features of the two files and comapre these for
similarity. However,I am confused as to what the main features of a
song are. I read somewhere that apparantly a song is made up of lots
of data in the low frequency range and lots of data in the high
frequency range, and that the low frequency data is the most
important. But I have not been able to confirm this at all.

Can anyone confirm what I have read, or suggest other answers?

Thank you very much

David Saunders

[Adrian Sanabria]

If you want to do a waveform/spectrogram analysis, you can use a free
program called Audacity, available online:

1. Download Audacity here: http://audacity.sourceforge.net/
2. Install it.
3. Import both files, and you can compare the waveform by eye.

This may or may not work depending on how sensitive the differences are.
Is this audio music, or something else, like voice you are trying to
compare?

If it is music, tempo differences may make it difficult to compare,
although there is a plugin for Audacity that will allow you to match the
tempos.

Differences in recording quality may also make things more difficult. If
the two audio files were recorded with different devices (microphones?),
differences in noise and static will make the waveforms look a bit
different, or a lot different, depending on the amount. Of course, there
are Audacity plugins that remove noise and static as well.

--Adrian

[andychap]

Hi David,

All computer data consists of ones and zero's. Open the audio files up
in note pad and compare them there.

I would just select the first couple of seconds of each file to compare
because the files are huge and take ages to load . Use the free
audacity program to copy short samples from each file.

There is another way to to examine the files along the lines of
inverting the phase of one of them and playing them together. One
should cancel out the other if they are identical. If they are not the
same then you should only be able to hear the differences. Maybe
someone else will come along who knows about that type of thing.
Good luck with the rest of your project.

Andy

[Jeff]

You might want to investigate the algorithm that the use to classify music
in cddb. Its open source, so you should be able to get it. It parses a
piece of music and reduces its 'signature' to a single (large) number
which goes into their database.

I have digitized old recordings and run the cddb algorithm on them and
discovered that it correctly identifies the piece based on a much newer
recording (not consistently, but occasionally). You'd have to no doubt
tweak it a bit to take into account pitch and tempo shifts.

-jeff