Audacity Forum

Whether built into Audacity or as an add-on like Nyquist or whatever:

Blind Source (or Signal) Separation: Can pull apart a number of sources into discrete output signals, convolved into a single mono signal. Can be used to separate signal and noise, whether pseudo-random (pink, brown, thermal, etc.) noise, or multiple similar sources (one voice among many -- the "cocktail party effect").

Continuous wavelet transform: A method of spectral analysis far beyond FFT. Can develop an accurate time/frequency plot from signal for doing time selective filtering, artifact detection, all manner of signal/noise separation.

Both are available as code of various flavors, most common being Matlab. Both can be used as the basis for subtractive s/n reduction, or simply separation so one can use that as a decision making tool for what other analysis or alteration techniques to apply. I can help with describing what they do and how, and can assist with existing code location and code author assistance/advice, but I can't code for squat. Rather I can, but if you need anything newer than Apple ][ machine level programming or more general that SPSS automation, you'd best look elsewhere.

"No matter what you're trying to do, it's good to remember that just as weeds are plants you don't want, noise is signal you don't want." -- Karl Pribram (1919 - 2015)

I can't code for squat.

That may be what's needed. Those tools are very desirable but if nobody can produce the actual code, then they may be "available on somebody else's program" for a while.

kozikowski wrote:

I can't code for squat.

That may be what's needed. Those tools are very desirable but if nobody can produce the actual code, then they may be "available on somebody else's program" for a while.

The code and executables for both exist, and have for 20 years. For Matlab and other analysis packages, as well as generic C++ and similar. Most is FOSS, some is even GNU/GPL, and there are Sourceforge projects for them. As for what to do with them, in a way to make it acceptable to be included in Audacity, if I could do that I'd be doing it rather than suggesting it in this forum.

drmcclainphd wrote:Continuous wavelet transform: A method of spectral analysis far beyond FFT. Can develop an accurate time/frequency plot from signal for doing time selective filtering, artifact detection, all manner of signal/noise separation.

A "transform" is a mathematical operation, just as addition, multiplication, convolution, DFT and so on are mathematical operations. I'm not clear about what you would want this to "do".

steve wrote:A "transform" is a mathematical operation, just as addition, multiplication, convolution, DFT and so on are mathematical operations. I'm not clear about what you would want this to "do".

https://en.wikipedia.org/wiki/Continuou ... _transform

Since wavelet transform has the ability to decompose complex information and patterns into elementary forms, it is commonly used in acoustics processing and pattern recognition.

Brian Davies' Click Repair uses wavelets to identify clicks, remove them, then reconstruct the signal.

-- Bill

billw58 wrote:https://en.wikipedia.org/wiki/Continuou ... _transform

OK, so a bit more complicated than addition or multiplication, but nevertheless it is still just a mathematical operation and in itself "does" nothing.

Audacity has a couple of FFT functions, which of themselves are not very interesting to users because they are just mathematical operations. What is of interest to end users is what Audacity "does" with those functions (namely "Plot Spectrum", track spectrogram view, Equalization and Noise Reduction effects).

billw58 wrote:Brian Davies' Click Repair uses wavelets to identify clicks, remove them, then reconstruct the signal.

Thanks Bill, yes that could be one application. It can also be used in granular synthesis, for extracting known signals from noise, for speech analysis and synthesis, for physical modelling of resonant systems and many other applications, but I'm still no clearer about what drmcclainphd wants Audacity to "do" with CWT.