Chris Cannam
|
638948269b
|
Remove Silent classification from BinClassifier: not only is it not very useful, it's misaligned with the other classifications because it doesn't account for lag, and so it can make those wrong
|
2022-06-14 13:59:17 +01:00 |
|
Chris Cannam
|
fb75aa36f6
|
Using the instantaneous ratio here seems more proper, but it causes audible interference when using tiny shift ratios such that the hop flips back and forth between two adjacent values. Let's try using the more stable effective ratio, and see if the window smoothing is sufficient
|
2022-06-14 10:01:11 +01:00 |
|
Chris Cannam
|
9dbf53026a
|
Fix failure to calculate offline stretch correctly when a zero-chunk region is found; provide alternate projection calculation method for R3 without stop-the-world phase resets
|
2022-06-13 17:16:03 +01:00 |
|
Chris Cannam
|
ac4072937e
|
Add process mode; start on key-frame map
|
2022-06-13 16:06:21 +01:00 |
|
Chris Cannam
|
90ad1274d8
|
Add R3 support to plugins
|
2022-06-13 11:49:04 +01:00 |
|
Chris Cannam
|
c7e4d9eb07
|
Provide option to shift formant independently of pitch
|
2022-06-13 10:39:13 +01:00 |
|
Chris Cannam
|
182e2b0e3b
|
Put binForFrequency/frequencyForBin in a common place
|
2022-06-13 10:08:05 +01:00 |
|
Chris Cannam
|
05fb611544
|
Experimental preKick to slightly reduce frame just before kick and boost kick accordingly
|
2022-06-13 09:40:26 +01:00 |
|
Chris Cannam
|
321a89e372
|
Fix absurd typo
|
2022-06-10 20:48:17 +01:00 |
|
Chris Cannam
|
9f913385de
|
Apply start skip. Still work to be done
|
2022-06-10 20:26:37 +01:00 |
|
Chris Cannam
|
a66b01446f
|
Pre-pad start of input; now need to un-pad start of output
|
2022-06-10 18:15:58 +01:00 |
|
Chris Cannam
|
2298b67869
|
Avoid recalculating window scale factor every time
|
2022-06-09 16:39:30 +01:00 |
|
Chris Cannam
|
14394141b0
|
Comment out debug
|
2022-06-09 16:00:27 +01:00 |
|
Chris Cannam
|
a7353d7161
|
Tidy
|
2022-06-09 15:39:06 +01:00 |
|
Chris Cannam
|
4a9574b7ee
|
Backed out changeset b1275ea1b6c8
This makes very little perceptible difference, possibly for the worse, and it is more complicated
|
2022-06-09 15:26:16 +01:00 |
|
Chris Cannam
|
4457247749
|
Synchronise frequency channel cutoffs across channels. Not the most elegant way to do this I think
|
2022-06-09 14:56:20 +01:00 |
|
Chris Cannam
|
ab284f0047
|
Rather than using trough picker, make the division frequencies drift downhill
|
2022-06-09 14:29:51 +01:00 |
|
Chris Cannam
|
70a7b6d688
|
Some (temporary) debug etc
|
2022-06-09 14:16:49 +01:00 |
|
Chris Cannam
|
95a1d6df25
|
Start to bring in unit tests
|
2022-06-08 09:57:12 +01:00 |
|
Chris Cannam
|
36f993b9a6
|
Reduce p level for trough picker. The behaviour here is not what I'd hoped
|
2022-06-07 14:13:24 +01:00 |
|
Chris Cannam
|
07b339df13
|
Try a peak p-factor of 1 for the previous peaks
|
2022-06-07 12:12:06 +01:00 |
|
Chris Cannam
|
8ed709d274
|
Don't run median filters for higher frequencies than we're going to use anyway
|
2022-06-07 11:05:50 +01:00 |
|
Chris Cannam
|
c31839ef93
|
Use fixed (faster) resampler settings in offline mode
|
2022-06-07 09:06:52 +01:00 |
|
Chris Cannam
|
ef2d39b3af
|
Or is this the right thing? I'm having trouble deciding
|
2022-06-06 17:50:22 +01:00 |
|
Chris Cannam
|
48174ded20
|
Ensure we have full magnitude range available in classify scale even when ratio has changed and the readahead isn't valid - necessary to avoid noise on ratio change when formant shifting
|
2022-06-06 17:44:22 +01:00 |
|
Chris Cannam
|
6ef77ef7aa
|
Tidy
|
2022-06-06 17:10:12 +01:00 |
|
Chris Cannam
|
100a7c0ba6
|
Go back to running formants only within the necessary range
|
2022-06-06 17:06:52 +01:00 |
|
Chris Cannam
|
4fb7b0ad47
|
And per-channel formants
|
2022-06-06 16:51:02 +01:00 |
|
Chris Cannam
|
ec7a2b1b51
|
Experiment with formant shifting before the peaks/phase work
|
2022-06-06 16:37:44 +01:00 |
|
Chris Cannam
|
9a7a977fa0
|
Tidy
|
2022-06-06 14:49:01 +01:00 |
|
Chris Cannam
|
0ceca73636
|
Settle on these values for the moment
|
2022-06-06 14:00:09 +01:00 |
|
Chris Cannam
|
f8abd07438
|
Tidy, and try without a high formant limit
|
2022-06-06 13:09:29 +01:00 |
|
Chris Cannam
|
baab6ae66e
|
Further formant experiments
|
2022-06-06 12:08:52 +01:00 |
|
Chris Cannam
|
2673983220
|
I think this check was only necessary because of our wrong handling of pitch scale
|
2022-05-27 15:29:28 +01:00 |
|
Chris Cannam
|
ce64122ffe
|
Wire up formant option; some tidying
|
2022-05-27 15:16:10 +01:00 |
|
Chris Cannam
|
7febc09dbc
|
Fix accidental time stretch coming along with pitch shift
|
2022-05-27 15:03:40 +01:00 |
|
Chris Cannam
|
f13d96a474
|
First cut at formant preservation
|
2022-05-27 14:58:42 +01:00 |
|
Chris Cannam
|
9379908982
|
Avoid excessive outhop when pitch shifting a long way
|
2022-05-27 11:34:35 +01:00 |
|
Chris Cannam
|
9ecc601a2c
|
Implement reset
|
2022-05-27 11:17:20 +01:00 |
|
Chris Cannam
|
580d28afd2
|
Correct the use of in/out hop vs prev in/out hop
|
2022-05-27 10:06:31 +01:00 |
|
Chris Cannam
|
08eeb17a98
|
Merge
|
2022-05-26 17:46:23 +01:00 |
|
Chris Cannam
|
84cd061e9d
|
Avoid duplicate bin in the different scales
|
2022-05-26 17:46:13 +01:00 |
|
Chris Cannam
|
2c23d52c94
|
Clarify
|
2022-05-26 17:42:45 +01:00 |
|
Chris Cannam
|
dd8d328ffb
|
Merge
|
2022-05-26 15:09:40 +01:00 |
|
Chris Cannam
|
83f2b7607b
|
An attempt to do the right thing when the hop changes - but this is not enough
|
2022-05-26 15:08:07 +01:00 |
|
Chris Cannam
|
fa004562f7
|
Tidy, and format comments
|
2022-05-25 16:25:03 +01:00 |
|
Chris Cannam
|
630a790ef8
|
Use atomics for inhop and ratio, and read inhop once at start of consume in case it changes within
|
2022-05-25 14:10:41 +01:00 |
|
Chris Cannam
|
eb017476ca
|
Avoid resetting m_prevOuthop on every set-ratio call (client may call this very repeatedly)
|
2022-05-25 13:51:23 +01:00 |
|
Chris Cannam
|
54515122b2
|
Implement resampling
|
2022-05-25 13:47:40 +01:00 |
|
Chris Cannam
|
f5b381e086
|
Pull out per-channel analysis and resynthesis functions
|
2022-05-25 11:26:16 +01:00 |
|