Simplify confidence metric
oops! also avoid memory leak
Thread safety
More tweaking with thresholds and confidences -- but I feel like this is going in the wrong direction
Merge
Change peak-to-next-peak ratio to a difference (because difference of peaks in log values don't vary with signal amplification)
OS/X build bits
Clumsily cut-n-paste interpolation into pitch tracker as well
Adjust hypothesis tolerances (i.e. note breaking algorithm)
A first crack at turning frequencies into notes
Introduce peak-to-second-peak ratio, which looks like a reasonable proxy for harmonic-ness. Use it to ascribe a confidence to estimates in the pitch tracker & rely on that to determine how many similar estimates make a satisfied hypothesis
Reap expired/rejected hypotheses from list; report an estimate as accepted only if the accepting hypothesis has been satisfied already (otherwise a different provisional hypothesis might become satisfied first)
Another small fix
Make it, at least broadly, work
Make compile; does not yet work
Closing unsuccessful anonymous branch
An experiment in mean-filtering proportion values -- not very successful
Partial updates to new method (but this doesn't currently build)
Start an alternative idea
Some work on accepting/rejecting peaks
Vertical filtering
Switch to inverse symmetric method. Forward difference worked nicely for some examples but very badly for others (octave errors)
Experimental tracker (on branch)
Add total and peak-proportion outputs
Change "relative to mean" to "relative to RMS", and add peak-to-RMS output. Add a mean-filter history. Remove essentially useless forward-power method (same as forward-magnitude with 2x factor). Refactor a bit
« Previous 1 2 3 4 Next » (51-75/76) Per page: 25, 50, 100
Also available in: Atom