[linux-audio-user] Note tuning and quantizer in audio files
alex_osiris at hotmail.com
Mon Jul 5 11:53:16 EDT 2004
Just to briefly explain these two nice techniques as a couple of you have
asked about them. They are both based on the same math, a kind of modified
fast fourier transform which enables the software to stretch or shrink time
without affecting pitch or sound timbre (the characteristic that enables us
to recognize an instrument).
Note tuning, only applicable to samples of monophonic instruments: the
software has a freq detector that will find the nearest note to what is
being played. It will find a starting time, end time and deviation for that.
Then it will apply this FFT to generate a sample equal in duration but in
tune. This is done by stretching or shrinking as appropriate first (this
does not change pitch), then re-sampling to length which changes the pitch
Quantizer: only used for samples of percussion AFAIK, but should be
applicable to other instruments with high attack like bass and others: first
step is for the software to identify the location of the notes (that's the
equivalent for MIDI events but this time in audio), by using a kind of peak
detector this time instead of the freq detector for the other technique.
Second step is to identify where the notes should exactly be time-wise. For
this, it's neccessary that the software knows about the timing of the song /
sample! (basically BPM and desired quantize resolution). Knowing this, it's
pretty straightforward to "shrink and stretch" before and after the note so
it falls exactly on time.
These two give best results when the notes are not too off. This is due to
this special FFT not being a perfect method to extract exactly everything
that's there, especially high frequencies. Different implementations of
these use different algorithms, all are based on a FFT, but some use
different techniques to mask certain side effects that appear on the
modified sample. Some implementations are very impressive I have to say, and
they can stretch or shrink to say two times or half with brilliant results.
All pretty impressive technology which has been around for a while now..
Horóscopo, tarot, numerología... Escucha lo que te dicen los astros.
More information about the Linux-audio-user