[LAD] [Source uploaded to GitHub] GuitarSynth

Fons Adriaensen fons at linuxaudio.org
Sun Apr 26 22:21:06 UTC 2015

On Sun, Apr 26, 2015 at 01:00:40PM -0400, Tim E. Real wrote:
> Fons do you have any insight into wavelets and how they might
> be better for lower latency pitch detection than FFT?

I don't think they will allow you to have better latency
if all other aspects of performance are kept equal. 

Don't expect magic from wavelets - they can't provide 
information that isn't there. What they allow you to do
is modify the tradeoff between resolution in time and
frequency.  But the product of the two is limited by
the 'uncertaintly principle' in all cases: better 
resolution in time means less resolution in frequency
and vice versa. 

There is no way to break this limit *unless* you make
assumptions about the signal to be detected and accept
failure if these are not valid. Human hearing seems to
do this. In the case of a guitar signal there are some
valid assumptions, e.g. the maximum number of notes
and the typical exponential decay of their envelope.

For the last ten years or so, a new technique called
'compressive sensing' has been a hot topic in DSP.
It is being used to do all sorts of things that at
first sight seem impossible, such as recovering a 
signal from random samples well below the Nyquist 
rate. It again depends on making certain assumptions
about the analysed signal, in this case that there
exists a 'sparse' representation of it.   

The whole theory behind CS is a bit counter-intuitive
but mathematically perfectly sound. I'm pretty sure
it will lead to some new ways to do pitch detection. 



A world of exhaustive, reliable metadata would be an utopia.
It's also a pipe-dream, founded on self-delusion, nerd hubris
and hysterically inflated market opportunities. (Cory Doctorow)

More information about the Linux-audio-dev mailing list