Alignment is now right in inverse transform, but the padding at end is still wrong
Fix bin ordering within octave
Inverse runs; now merely needs to run correctly
More work on inverse
A bit more on inverse
Bit more on inverse
Rearrange classes so the basic ConstantQ produces a complex output, and CQSpectrogram (formerly CQInterpolated) is required if you want a real output
More toward inverse CQ
Start on C++ inverse
Fix compiler warning
Reconstruction now appears to be (roughly!) correct in the yeti code
Inverse now produces a (wrong) output
Add small program to run some experiments
Another step in icqt
A bit more on icqt
More on tests and inverse stuff
Start on inverse code, fix up forward interface a bit
Backed out changeset 3393a2898a1d
Try chaining the resamplers
Avoid compiler warnings
Linear interpolation fixed
Interpolation for CQ. But this is wrong, it is interpolating between points in the nearest full-height columns even where there is a nearer point in the not-full-height ones
vector multiply now takes a list
Ant build file
Support returning a magnitude spectrum (dense) etc
Vamp category and RDF files
minor tidy
Copyrights, licence
Allow wrong max in alternate columns where expected max would be in lower octave (which is not present)
Plot failing test matrices
Add little program to plot cq of an audio file
Pull out tests
Comment out some debug
Just an import
Build fixes
Merge with pulled changes
Start on automated tests
Put spectrogram the "right" way up, add bin labels
Fix min frequency calculation in accessor, add bin frequency accessor
Single Makefile, build Vamp by default
Use MIDI pitch values, as in QM Vamp Plugins implementation
Correct output latency
Further tidying, compensate for latency in Vamp plugin
Simplify
Further latency adjustment
Remove debug out
We shouldn't need to compensate for max(latency)+max(drop) but for max(latency+drop); also the padding can be rounded to hop rather than fft size I think?
Towards handling the hop alignment (which in the MATLAB is handled on output) at the input end
concat (map ... ) -> concatMap
Make additional latency at start a multiple of the big blocksize
Start to address some padding and latency issues
Some work on alignment
Update to revised api
Use revised framer api
More debug testing
Return bin frequencies, for reference; some padding
Use mat.print
Adjust test case to match matlab example on wiki; print out results
Pull out test program
Get remaining blocks from end of processing; fix some compile warnings, etc
Add Vamp plugin for testing with
More on reconstructing CQ shape. This is not always correct yet, though
Delete the kernel!
Much of the actual CQ processing logic
Don't forget top octave in buffer list
Start on CQ process implementation
Stub class
Ah, the code was right, the test was wrong
Almost finish normalising kernel and storing it sparsely
Minor updates
bf -> vec, matrix concat rename
Keep the kernel dense for the moment, and use std::complex
Fix frequency calculation in cpp, and some diagnostics
Calculate (so far unnormalised) CQ kernel
Restore reshape into atom columns; drop atoms for alignment (not working, we may be compensating in the wrong direction)
Return only the actual results (i.e. space with zeros rather than duplicates)
Build assembled blocks
More on assembling blocks
Start trying to assemble blocks as they arrive
Print out results (to go to file to load into e.g. SV)
Towards shuffling the cq cells into the right arrangement for return
Reshape result vectors
Comments
Return the atom count & spacing as well
Start on code to apply the transform
Make into a function!
Normalisation -- now it really matches the toolbox
Use symmetric window -- kernel now matches the MATLAB one
Sparse etc
Return sparse complex matrix
Minor fixes
Generate the actual kernels
Another minor few sums...
Initial kernel parameters
Directory structure, paper, matlab etc
« Previous 1 2 (101-197/197) Per page: 25, 50, 100
Also available in: Atom