14 Mar
2004
14 Mar
'04
11:03 a.m.
hello list,
a colleague of mine seems to be extremely interested in performing
exaggerated time stretching of audio files. i wrote a csound instrument
based on the new streaming phase vocoding opcodes by richard dobson.
adjusting window size, it can perform passable time stretching up to 6 or 8
times the duration of the original.
but this colleague would like to have much more than that. he told me in in
a visit to cologne he used a POD system, where he could stretch sounds up to
400 or 500 times...
my questions are:
- which linux applications can stretch sounds with good quality up to let's
say at least 20 times?
- is there anything like POD for linux? anyone knows how POD does the
time-stretching? granular synthesis?
thanks as always.
lj
--
14 Mar
14 Mar
12:30 p.m.
luis jure wrote:
- which linux applications can stretch sounds with
good quality up to let's
say at least 20 times?
Ardour [pretty good engine for this]
Pure Data [look for Frank Barknecht's Syncgrain external, or my own
Particle Chamber abstraction, for a synchronus/asynchronous granular
synthesis approach.]
there are some commmand-line libraries as well, can't remember the
names. check the archives of this list for "time stretching" and you
will see a long discussion on the topic.
d.
--
derek holzer ::: http://www.umatic.nl
---Oblique Strategy # 39:
"Cut a vital connection"
12:39 p.m.
Hallo,
luis jure hat gesagt: // luis jure wrote:
my questions are:
- which linux applications can stretch sounds with good quality up to let's
say at least 20 times?
- is there anything like POD for linux? anyone knows how POD does the
time-stretching? granular synthesis?
I'd say, that every granular synthesizer can stretch your sound to
infinity. Phase vocoding should be able to do this as well. One of
the fun applications of grain synthesis is to "stop" right into a
sound, so that the movement slows down to a halt.
Here is a quick'n'dirty example - without any finetuning of parameters
for cleaner sound - using the snycgrain~ external in Pd:
http://footils.org/snd/slow.ogg
It's first playing some bars in original time, then slows to a stop,
then accelerates again up to doublespeed.
ciao
--
Frank Barknecht _ ______footils.org__
15 Mar
15 Mar
9:28 a.m.
el Sun, 14 Mar 2004 18:33:31 +0100
Frank Barknecht <fbar(a)footils.org> escribió:
Here is a quick'n'dirty example - without any
finetuning of parameters
for cleaner sound - using the snycgrain~ external in Pd:
http://footils.org/snd/slow.ogg
thanks for your answer, it's a very interesting example, indeed. but i'm
afraid i don't use PD... :-( i'm just an old-fashioned csound guy...
yes, i know that with granular synthesis you can in theory stretch any sound
infinitely, but i never could make it sound clean and well. i'm trying with
the fog opcode, but i'm not satisfied with the results. the phase vocoding
approach sounds much better, with long analysis windows and high number of
superpositions i get passable results up to about 15 times.
i'll also try ardour, as suggested.
best,
lj
3:26 p.m.
On Sun, Mar 14, 2004 at 06:33:31PM +0100, Frank Barknecht wrote:
Hallo,
luis jure hat gesagt: // luis jure wrote:
What is granular synthesis? For the purposes of time stretching or
compressing sound, how does it compare to the way Erik's SRC or
SoundTouch work?
Thanks,
Eric Rz.
my questions are:
- which linux applications can stretch sounds with good quality up to let's
say at least 20 times?
- is there anything like POD for linux? anyone knows how POD does the
time-stretching? granular synthesis?
I'd say, that every granular synthesizer can stretch your sound to
infinity. Phase vocoding should be able to do this as well. One of
the fun applications of grain synthesis is to "stop" right into a
sound, so that the movement slows down to a halt. 
3:43 p.m.
On Mon, 15 Mar 2004 15:21:01 -0500
Eric Dantan Rzewnicki <rzewnickie(a)rfa.org> wrote:
What is granular synthesis?
Google might help for a better answer, but basically yo chop the
sound into short segments and then repeat/drop segments as
requires.
For the purposes of time stretching or
compressing sound, how does it compare to the way Erik's SRC or
SoundTouch work?
SRC is a sample rate converter and hence cannot modify the length
without modifying pitch.
SoundTouch is a true pithc independant time stretcher.
Erik
--
+-----------------------------------------------------------+
Erik de Castro Lopo nospam(a)mega-nerd.com (Yes it's valid)
+-----------------------------------------------------------+
When aiming for the common denominator, be prepared for the
occasional division by zero.
4:26 p.m.
Hallo,
Erik de Castro Lopo hat gesagt: // Erik de Castro Lopo wrote:
On Mon, 15 Mar 2004 15:21:01 -0500
Eric Dantan Rzewnicki <rzewnickie(a)rfa.org> wrote:
I think, Soundtouch uses a Fourier-based approach. Granular synthesis
isn't really the best method of clean time stretching, I just
mentioned it as one example of a technique, that can be used to do
extreme streches.
BTW, one of the beautifullest examples of extreme stretching isn't
Jane Fonda, but the "Beetstretch" project of Leif Inge:
http://www.notam02.no/9/ who stretched Beethoven's 9th Symphony -
which originally is about 1 hour - to be 24 hours long. The
audio-files are several gigabytes large, an obvious use case for
Bittorrent.
Maybe Kjetil would elaborate a bit, how this was done, as I read that
he was involved in the Beetstretch?
ciao
--
Frank Barknecht _ ______footils.org__
For the purposes of time stretching or
compressing sound, how does it compare to the way Erik's SRC or
SoundTouch work?
SoundTouch is a true pithc independant time stretcher. 
4:53 p.m.
On Mon, Mar 15, 2004 at 10:21:00 +0100, Frank Barknecht wrote:
Hallo,
Erik de Castro Lopo hat gesagt: // Erik de Castro Lopo wrote:
Soundtouch uses WSOLA, which is similar to granular synthesis, its not
fequency domain like FFT or wavelets.
- Steve
On Mon, 15 Mar 2004 15:21:01 -0500
Eric Dantan Rzewnicki <rzewnickie(a)rfa.org> wrote:
I think, Soundtouch uses a Fourier-based approach. Granular synthesis
isn't really the best method of clean time stretching, I just
mentioned it as one example of a technique, that can be used to do
extreme streches. For the purposes of time stretching or
compressing sound, how does it compare to the way Erik's SRC or
SoundTouch work?
SoundTouch is a true pithc independant time stretcher.
5:18 p.m.
On Mon, Mar 15, 2004 at 09:48:05PM +0000, Steve Harris wrote:
On Mon, Mar 15, 2004 at 10:21:00 +0100, Frank
Barknecht wrote:
Waveform Similarity Overlap Add?
http://bmrc.berkeley.edu/courseware/cs298/spring00/Ponceleon/Dulce/index4.h…
(not that i can claim to understand that picture ... :-/ )
-Eric Rz.
Hallo,
Erik de Castro Lopo hat gesagt: // Erik de Castro Lopo wrote:
Soundtouch uses WSOLA, which is similar to granular synthesis, its not
fequency domain like FFT or wavelets.
- Steve On Mon, 15 Mar 2004 15:21:01 -0500
Eric Dantan Rzewnicki <rzewnickie(a)rfa.org> wrote:
I think, Soundtouch uses a Fourier-based approach. Granular synthesis
isn't really the best method of clean time stretching, I just
mentioned it as one example of a technique, that can be used to do
extreme streches. For the purposes of time stretching or
compressing sound, how does it compare to the way Erik's SRC or
SoundTouch work?
SoundTouch is a true pithc independant time stretcher.
5:51 p.m.
On Mon, Mar 15, 2004 at 05:13:13 -0500, Eric Dantan Rzewnicki wrote:
On Mon, Mar 15, 2004 at 09:48:05PM +0000, Steve Harris
wrote:
Yep, thats the one. Its just about slicing up the waveform into sections
and trying to paste them back together in overlapping ways, but so that
the features in the original sections line up.
- Steve
On Mon, Mar 15, 2004 at 10:21:00 +0100, Frank
Barknecht wrote:
Waveform Similarity Overlap Add?
http://bmrc.berkeley.edu/courseware/cs298/spring00/Ponceleon/Dulce/index4.h… Hallo,
Erik de Castro Lopo hat gesagt: // Erik de Castro Lopo wrote:
Soundtouch uses WSOLA, which is similar to granular synthesis, its not
fequency domain like FFT or wavelets.
- Steve On Mon, 15 Mar 2004 15:21:01 -0500
Eric Dantan Rzewnicki <rzewnickie(a)rfa.org> wrote:
> For the purposes of time stretching or
> compressing sound, how does it compare to the way Erik's SRC or
> SoundTouch work?
SoundTouch is a true pithc independant time stretcher.
I think, Soundtouch uses a Fourier-based approach. Granular synthesis
isn't really the best method of clean time stretching, I just
mentioned it as one example of a technique, that can be used to do
extreme streches.
6:51 p.m.
Doing some quick experiments with Soundtouch, I hear that it probably
isn't going to work for "extreme" timestretching. Even doubling the
length of a 4 minute soundfile results in *very* noticeable artifacts...
a kind of stuttering effect as the samples get pulled apart. You won't
be making a 24 hour Beethoven stretch with this one...
Seems like an FFT-based approach, or a very good synchronous
granulation, might be better. Pure Data has a nice phase vocoder demo
patch in its documentation. You'll have to fiddle with it to get stereo,
of course. But hey, it's only a help patch! ;-)
Other option is the Synchgrain external for PD, as I might have
mentioned earlier. I haven't tested the limits of it yet, but I know
that it sounded smooth at pretty slow playbacks.
d.
--
derek holzer ::: http://www.umatic.nl
---Oblique Strategy # 173:
"Voice your suspicions"
11:14 p.m.
I've done some pretty extreme stretching/compressing of time with
ecasound's -ei effect (labeled as a pitch shifter). The default builtin
implementation (linear interpolation, I think?) will do arbitrary
percentages up or down. I don't have a great critical ear for sound
quality (things tend to just sound different in different ways rather
than better or worse) but it seems fine to me and works. If you compile
ecasound with SRC support the shift is limited to 12x in either
direction. But, as Steve pointed out to me on #lad last week(?) you can
do multiple iterations to get greater shifts/stretches. Using SRC the
resulting sound is somehow smoother sounding than with the internal
ecasound version.
-Eric Rz.
derek holzer wrote:
Doing some quick experiments with Soundtouch, I hear
that it probably
isn't going to work for "extreme" timestretching. Even doubling the
length of a 4 minute soundfile results in *very* noticeable artifacts...
a kind of stuttering effect as the samples get pulled apart. You won't
be making a 24 hour Beethoven stretch with this one...
Seems like an FFT-based approach, or a very good synchronous
granulation, might be better. Pure Data has a nice phase vocoder demo
patch in its documentation. You'll have to fiddle with it to get stereo,
of course. But hey, it's only a help patch! ;-)
Other option is the Synchgrain external for PD, as I might have
mentioned earlier. I haven't tested the limits of it yet, but I know
that it sounded smooth at pretty slow playbacks.
d.
16 Mar
16 Mar
1:31 a.m.
On Mon, 15 Mar 2004 23:09:22 -0500
Eric Dantan Rzewnicki <eric(a)zhevny.com> wrote:
<snip>
If you compile
ecasound with SRC support the shift is limited to 12x in either
direction.
That was true for SRC up to and including 0.0.15. For version 0.1.0
released a couple of days ago, that has changed to +/- 256.
Also, be aware that this is not a pitch shift effect, because the
durartion is not constant. The effect is more accurately described
as a vari-speed effect.
But, as Steve pointed out to me on #lad last week(?)
you can
do multiple iterations to get greater shifts/stretches. Using SRC the
resulting sound is somehow smoother sounding than with the internal
ecasound version.
If the internal ecasound on is linear or polynomial interpolation,
then that would make a lot of sense.
Erik
--
+-----------------------------------------------------------+
Erik de Castro Lopo nospam(a)mega-nerd.com (Yes it's valid)
+-----------------------------------------------------------+
"Microsoft, and other companies with shoddy security, ....."
-- Bruce Schneier, cryto-guru, to a US Senate committee.
12:11 p.m.
On Tue, Mar 16, 2004 at 05:25:30PM +1100, Erik de Castro Lopo wrote:
On Mon, 15 Mar 2004 23:09:22 -0500
Eric Dantan Rzewnicki <eric(a)zhevny.com> wrote:
That's excellent news. :) thank you very much. I'll rebuild my ecasound
later this week. http://www.mega-nerd.com/SRC/index.html still refers to
the 12x limit.
If you compile
ecasound with SRC support the shift is limited to 12x in either
direction.
That was true for SRC up to and including 0.0.15. For version 0.1.0
released a couple of days ago, that has changed to +/- 256. Also, be aware that this is not a pitch shift effect,
because the
durartion is not constant. The effect is more accurately described
as a vari-speed effect.
I see. That makes sense.
If the internal ecasound on is linear or polynomial
interpolation,
then that would make a lot of sense.
What does SRC use?
-Eric Rz.
7703
days inactive
7705
days old
linux-audio-user@lists.linuxaudio.org
14 comments
7 participants
participants (7)
-
derek holzer -
Eric Dantan Rzewnicki -
Eric Dantan Rzewnicki -
Erik de Castro Lopo -
Frank Barknecht -
luis jure -
Steve Harris