time stretching of samples #149

bgold-cosmos · 2019-10-10T15:27:25Z

This implements time-scaling for sample playback using an overlap-add (OLA) algorithm.

I tweaked the code with speed and flexibility as priorities rather than output sound quality, but for modest scaling amounts (0.8 to 1.3 or so) it usually sounds fairly good. I'm not aware of any method that will consistently timescale transient-only sounds (most drums) in a "nice" way.

The timescaling should play well with speed, begin, end and similar parameters, but I'm sure I haven't tested every possible combination.

Examples of usage in Tidal would be

timescale = pF "timescale"

d1 $ chop 4 $ s "rave" # timescale "[1 4] 2" # cps 0.5

d1 $ s "metal*4" # cps 1
 # up ("c g6 c c6" + "<c ef g c>" - 12)
 # timescale "1 3 1 2"

Sync to master

Release to master, in sync with tidal v1.1.0

Small fixes to master (LATEST)

initial commit, mostly seems to work but the parameter name(s) and implementation details may change

telephon · 2019-10-10T17:28:33Z

just to be sure: have you checked the PitchShift UGen for comparison? Your implementation has the advantage that it could be combined with the https://github.com/musikinformatik/WavesetsEvent Quark, that sounds much better to my ear than overlap add.

In the end, of course we shouldn't load the standard player with extra things, but add a switch when timestretch is set, to save electricity.

bgold-cosmos · 2019-10-18T16:38:21Z

Whenever I've tried the PitchShift UGen I've gotten really choppy results regardless of windowSize. However, I should check again to do a sound comparison, and I'll take a look at the Wavesets stuff.

I wasn't sure how best to conditionally apply the effect. My understanding is it's not really possible to have true branching in a SynthDef, so would there have to be a whole new set of dirt_sample_timescale_%_% SynthDefs that would get called when the event's ~timeScale is non-nil? I tried to keep the extra computation to a minimum - all it really is doing is making a pair of phase indices for BufRd rather than a single one, and doing a little extra modulation on them so that they're not quite simple LFSaws. I haven't rigorously compared CPU usage but it didn't seem to add much.

…into stretch

…SuperDirt into stretch" This reverts commit 1d6ce85, reversing changes made to 563cda1.

telephon · 2019-10-24T13:37:18Z

Why don't you implement it as an effect?

bgold-cosmos · 2019-10-28T14:48:34Z

The current version can compress time too, so to do that as an effect would require looking into the future. I don't know how to do that except for samples that've already been loaded. Also, I think working with the sample indexing will be much more efficient, it only ever has two overlapping windows --- whereas as an effect I think it would need to be implemented using a network of staggered delays, some potentially with a very long delay time for large stretches and long samples. Might be interesting to try both, though...

bgold-cosmos · 2019-10-28T14:49:09Z

The current version can compress time too, so to do that as an effect would require looking into the future. I don't know how to do that except for samples that've already been loaded. Also, I think working with the sample indexing will be much more efficient, it only ever has two overlapping windows --- whereas as an effect I think it would need to be implemented using a network of staggered delays, some potentially with a very long delay time for large stretches and long samples. Might be interesting to try both, though...

…

On Oct 24, 2019, at 8:37 AM, Julian Rohrhuber ***@***.***> wrote: Why don't you implement it as an effect? — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#149?email_source=notifications&email_token=AB7AIZWKJJLHAJBYJ4575H3QQGQJBA5CNFSM4I7ORIPKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOECFBVOA#issuecomment-545921720>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AB7AIZV5LKKQTUVV6GGUZSTQQGQJBANCNFSM4I7ORIPA>.

The time-scaling effect is now implemented as an alternate sample playback SynthDef that is only called if the `timescale` parameter is used from Tidal.

bgold-cosmos · 2019-11-22T21:25:15Z

OK, the timescaling has now been moved to a separate SynthDef so no extra work is done if it isn't used.

telephon

The extra complications introduced are worth it, I think! Thank you.

synths/core-modules.scd

synths/core-synths.scd

telephon · 2019-11-23T12:29:44Z

hacks/looper.scd

-once $ sound "fooloopReset"
+-- reset buffers. But do only one time (d1 $ silence),
+-- since d1 $ once $ sound "fooloopReset" does not work at my installation (1.4.3)
+d1 $ sound "fooloopReset"


Maybe my installation is ambiguous here?

I accidentally merged in something from the develop branch here, and I thought I undid it but apparently not. I didn't make this change.

classes/DirtEvent.sc

telephon · 2019-12-21T12:16:26Z

synths/core-synths.scd

@@ -68,13 +68,13 @@ live coding them requires that you have your SuperDirt instance in an environmen

 			numFrames = BufFrames.ir(bufnum);
 			windowSize = numFrames * timescale / 29.0;
-			windowSize = windowSize.clip(1000.0*timescale, 4410.0);
+			windowSize = windowSize.clip(1000*timescale, 4410);


Will that work on all sample rates? If not, replace 4410 by 0.1 * SampleRate.ir).

telephon · 2019-12-21T12:17:12Z

btw. sorry, I had overlooked your changes.

telephon · 2019-12-21T12:26:06Z

synths/core-modules.scd

-					out: ~out
-				])
+				if(~timescale.notNil) {
+					dirtEvent.sendSynth(~stretchInstrument, [


I'd change this to:

var instrument, args; ... args = [ ... ]; // the shorter arg list if(~timescale.notNil) { instrument = ~stretchInstrument; args = args.add(\timescale).add(~timescale) } { instrument = ~instrument }; dirtEvent.sendSynth(instrument, args); …

Then you don't need the same list twice.

telephon · 2019-12-21T12:26:53Z

synths/core-modules.scd

+						out: ~out
+					])
+				} {
+					// argumets could be omitted using getMsgFunc, but for making it easier to understand, we write them out


typo: argumets –> arguments

yaxu · 2021-01-23T15:13:23Z

Shall we make these smallish changes and @bgold-cosmos ?

bgold-cosmos · 2021-01-23T19:30:36Z

I'll refresh my memory on this and look into getting it updated. A lot of things got put on pause during 2020. I suspect given all the SuperDirt changes lately it'll be much cleaner to just redo this as a new PR - the semester is just starting Monday here but if I don't do something within a week or so don't feel bad about pinging me again.

telephon · 2021-01-23T20:34:39Z

I think it won't be hard to resolve any conflicts.

bgold-cosmos · 2021-02-01T15:43:06Z

I've redone this, see #229

yaxu and others added 14 commits April 25, 2019 21:59

Merge pull request musikinformatik#126 from musikinformatik/develop

02071f9

Sync to master

Merge branch 'master' of https://github.com/musikinformatik/SuperDirt

09cb77b

Merge branch 'develop' of https://github.com/musikinformatik/SuperDirt

99a88f1

Merge pull request musikinformatik#136 from musikinformatik/develop

61cf7e6

Release to master, in sync with tidal v1.1.0

Merge branch 'master' of https://github.com/musikinformatik/SuperDirt

913dc41

Merge branch 'develop' of https://github.com/musikinformatik/SuperDirt

343fae0

Merge pull request musikinformatik#144 from musikinformatik/develop

02c987c

Small fixes to master (LATEST)

Merge branch 'develop' of https://github.com/musikinformatik/SuperDirt

dfdda02

Merge branch 'master' of https://github.com/musikinformatik/SuperDirt

a34c500

Merge branch 'develop' of https://github.com/musikinformatik/SuperDirt

f305a25

Merge branch 'develop' of https://github.com/musikinformatik/SuperDirt

9530ad2

time stretching of samples

93b67d0

initial commit, mostly seems to work but the parameter name(s) and implementation details may change

tweak windowing and fix default stretch

ff93e82

rename to timescale, streamline calculations

563cda1

bgold-cosmos mentioned this pull request Oct 10, 2019

Timestretching? #146

Closed

bgold-cosmos added 2 commits October 22, 2019 09:30

Merge branch 'develop' of https://github.com/musikinformatik/SuperDirt …

1d6ce85

…into stretch

Revert "Merge branch 'develop' of https://github.com/musikinformatik/…

9a29e2a

…SuperDirt into stretch" This reverts commit 1d6ce85, reversing changes made to 563cda1.

Rework timescale as separate sample player

d654d21

The time-scaling effect is now implemented as an alternate sample playback SynthDef that is only called if the `timescale` parameter is used from Tidal.

telephon requested changes Nov 23, 2019

View reviewed changes

fixes as requested

11287b8

telephon reviewed Dec 21, 2019

View reviewed changes

telephon requested changes Dec 21, 2019

View reviewed changes

bgold-cosmos mentioned this pull request Feb 1, 2021

Stretch v2 #229

Merged

bgold-cosmos closed this Feb 1, 2021

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

time stretching of samples #149

time stretching of samples #149

bgold-cosmos commented Oct 10, 2019

telephon commented Oct 10, 2019

bgold-cosmos commented Oct 18, 2019

telephon commented Oct 24, 2019

bgold-cosmos commented Oct 28, 2019

bgold-cosmos commented Oct 28, 2019 via email

bgold-cosmos commented Nov 22, 2019

telephon left a comment

telephon Nov 23, 2019

bgold-cosmos Nov 25, 2019 •

edited

Loading

telephon Dec 21, 2019

telephon commented Dec 21, 2019

telephon Dec 21, 2019

telephon Dec 21, 2019

yaxu commented Jan 23, 2021

bgold-cosmos commented Jan 23, 2021

telephon commented Jan 23, 2021

bgold-cosmos commented Feb 1, 2021

time stretching of samples #149

time stretching of samples #149

Conversation

bgold-cosmos commented Oct 10, 2019

telephon commented Oct 10, 2019

bgold-cosmos commented Oct 18, 2019

telephon commented Oct 24, 2019

bgold-cosmos commented Oct 28, 2019

bgold-cosmos commented Oct 28, 2019 via email

bgold-cosmos commented Nov 22, 2019

telephon left a comment

Choose a reason for hiding this comment

telephon Nov 23, 2019

Choose a reason for hiding this comment

bgold-cosmos Nov 25, 2019 • edited Loading

Choose a reason for hiding this comment

telephon Dec 21, 2019

Choose a reason for hiding this comment

telephon commented Dec 21, 2019

telephon Dec 21, 2019

Choose a reason for hiding this comment

telephon Dec 21, 2019

Choose a reason for hiding this comment

yaxu commented Jan 23, 2021

bgold-cosmos commented Jan 23, 2021

telephon commented Jan 23, 2021

bgold-cosmos commented Feb 1, 2021

bgold-cosmos Nov 25, 2019 •

edited

Loading