DFT - how to recognize a wave

[ThomasT]

Hi!

A more academic and mathematical question:

I've got some little pieces of 24bit wave-files. Ca. 15ms long.
And I've got some data garbarge.
And I don't know where the audio data begins. With 24bit I have 3
possibilities to start with the right byte.

I want to write a program that can tell which is what.
Assuming that data garbage resuls in something like white noise I
tried to make a DTF from this (possible) audio pieces. But I got a lot
of high frequencies, too. So there ist no easy way to make an
relatively good guess.

I know in theorie there it nearly impossible, but it's normal music
and voices. Nothing spectacular. So there should be a way to be 80-90%
correct. Optical (waveform) data garbage and audio data are different.
But how to tell the computer?

Any Ideas?

Thomas Thiele

[Mark Tranchant]

ThomasT wrote:

> I've got some little pieces of 24bit wave-files. Ca. 15ms long.
> And I've got some data garbarge.
> And I don't know where the audio data begins. With 24bit I have 3
> possibilities to start with the right byte.

Interesting. I'd get the computer to work out the RMS difference between
the three bytes in adjacent samples. Assuming low levels of near-Nyquist
frequencies in the content, the set with the lowest differences averaged
over the sample should be the most significant byte of each sample.

Eurch. That's not clear at all, is it? What I mean is:

A1 -> A2 -> A3 etc
B1 -> B2 -> B3 etc
C1 -> C2 -> C3 etc

where A, B and C are the three bytes of the sample (only we don't know if
it's ABC, BCA or CAB) and 1, 2, 3 etc are the samples.

If the RMS differences between the B-samples, for example, turns out to be
smallest, I'd guess that B is the MSB and the order is BCA.

--
Mark.
http://tranchant.plus.com/