Natural Science Forum / Physics / Acoustics / July 2007

Hi,

I have a poor quality recording of a mono-tonic drum rhythm (a
recording where no two notes occur simultaneously).

I wish to convert this recording to midi and then to use samples (high
quality) to reconstruct the recording.

So far, I have written an attack algorithm which establishes where
each note begins. This works fine as far I can currently tell.
Therefore I now have a midi track of the drum rhythm, but the loudness
of each note is the same.

I have also produced an audio file, 128 seconds long, with a single
note from my drum sample library occuring every 1 second. Each note
rises in amplituded during the recording. This is achieved by
selecting midi loudness values of 0 to 127.

I now wish to measure the power of each note in the origional
recording as well as those in the sample recording. That way, I can
map a note in the origional recording with one in the samples
recording, such that they have the same loudness. This in turn allows
me to establish the midi loudness value for each note.

I wish to establish all this using an fft, which I have also used for
my attack detection algorithm. Would I be correct in saying that it
would be enough to establish the power at the attack for my purposes?
That is, to calculate the power at, and immediately after, the attack
point for each note. And to then match notes in each recording based
on this power value. My fft window is 2048 bins in length with a
44100Hz sampling rate.

Any suggestions?

Thanks in advance,

Barry.