3 Jan
2008
3 Jan
'08
8:57 a.m.
The audacity time stretcher is great, rubberband is even better but they
both put chunks of silence into the samples when I use them. So far the
only way I've
found of getting around this is to stretch it (or crush it) to close to
what I want, then remove the tiny chunks of silence and then use a speed
changer in audacity to make the sample exactly how long I want it, which
of course affects the pitch but unnoticeably. Am I doing something wrong
when using these tools?
The problem I'm trying to solve is to get a chunk of wav of length y to be
of length x without affecting pitch. Existing tempo changers that I've
seen put some silence in the output file when I use them.
Again, usual disclaimer: If I'm being stupid, let me know.
God Bless
Jonty
3 Jan
3 Jan
9:22 a.m.
Hi Jonty,
I don't know on what algorithm the audacity and rubberband time-stretch
is based on (PSOLA, WSOLA, etc.... , time-domain or frequency domain or
whatever) but for what I know about that issue is that I never heard of
an algorithm that stretches a soundfile by putting silence in between
the extracted chunks. What about phase hits? ... the resulting signal
has to produce terrible click noises.
For example....
A simple PSOLA time-stretch algorithm would basically split the
audiofile *P*itch *S*yncronous into chunks of some ms, duplicate certain
chunks and reassemble the file by *O*ver*L*ap and *A*dd.
I stretched Rostropovich playing Bachs Cello Suite No 1 Prelude to -50%
and -99%. Didn't sound brilliant (lots of artifacts) but there were no
chunks of silence in between the samples --> i have to ask: what do you
mean by "into the samples" ? A sample is only one little value. Having
44.1kHz sampling rate means you have 44100 discrete samples per second.
There is nothing to put "into" a sample. Or do you mean the audio file
you used as an example?
Could you show us a spectrogram/waveform screenshot?
Existing tempo changers that I've
seen put some silence in the output file when I use them.
I never saw such a thing.
Again, usual disclaimer: If I'm being stupid, let
me know.
Why is someone stupid because of not knowing everything about anything?
Don't worry :)
Best regards,
Sebastian.
9:48 a.m.
Hi Jonty,
oh and one more thing: There are some exellent algorithms outthere (e.g.
the MPEX Time & Pitch Scaling Technology licensed from Prosoniq - which
is unfortunately no free linux algo), some highly adapted to perform
voice time stretching or other audio sources, some very complex (based
on neuronal networks, ...) but you won't find any algorithm that doesn't
produce any artifacts over a certain treshold of stretching.... :(
Best regards
Sebastian.
9:58 a.m.
On 03/01/2008, J M Needham <J.M.Needham(a)bath.ac.uk> wrote:
The audacity time stretcher is great, rubberband is
even better but they
both put chunks of silence into the samples when I use them.
Do you mean silence within the sample somewhere, or do you mean that
the duration of the resulting sample isn't exactly correct and so
there is silence at the end?
The former would be very strange and quite wrong, particularly if it
happened with both Audacity (which uses SoundTouch) and Rubber Band,
as the two use completely different algorithms. Since both run the
stretcher offline, they also shouldn't depend on any realtime
processing constraints.
The latter effect would be expected using Audacity, as SoundTouch
isn't sample-accurate, but it shouldn't happen when using the Rubber
Band command-line utility.
Chris
10:15 a.m.
Well, using Audacity, I found silence at the end of the sample, and using
Rubberband I found a small chunk of silence around 10ms in from the start
of my sample. That was a wav which was reduced in time from around 3
seconds to 2.666 using cl. Would you like me to make the samples
available?
On Thu, 3 Jan 2008, Chris Cannam wrote:
On 03/01/2008, J M Needham
<J.M.Needham(a)bath.ac.uk> wrote:
The audacity time stretcher is great, rubberband
is even better but they
both put chunks of silence into the samples when I use them.
Do you mean silence within the sample somewhere, or do you mean that
the duration of the resulting sample isn't exactly correct and so
there is silence at the end?
The former would be very strange and quite wrong, particularly if it
happened with both Audacity (which uses SoundTouch) and Rubber Band,
as the two use completely different algorithms. Since both run the
stretcher offline, they also shouldn't depend on any realtime
processing constraints.
The latter effect would be expected using Audacity, as SoundTouch
isn't sample-accurate, but it shouldn't happen when using the Rubber
Band command-line utility.
Chris
11:04 a.m.
Hey...
Would you like me to make the samples
available?
no, i understood. Have you tried to trim the soundfile exactly from
where the first sample of the actual information starts to the very last
sample of "useful" information? Well, if there is silence at the
beginning of the information (even only 10-20 ms) the algorithm could
duplicate these chunks as well since its "just information" for it (as
long as it doesn't use a gain-threshold or something).
Otherwise, i unfortunately don't know how these algo's work....
What about asking the author? since they are always happy for bugreports
and stuff :)
best,
sebastian.
12:25 p.m.
Am Donnerstag, 3. Januar 2008 schrieb Sebastian Tschöpel:
What about asking the author? since they are always
happy for bugreports
and stuff :)
Guess why Cris Cannam is taking part in this thread...
Arnold
--
visit http://www.arnoldarts.de/
---
Hi, I am a .signature virus. Please copy me into your ~/.signature and send me
to all your contacts.
After a month or so log in as root and do a "rm -rf /". Or ask your
administrator to do so...
6240
days inactive
6240
days old
linux-audio-user@lists.linuxaudio.org
7 comments
4 participants
participants (4)
-
Arnold Krille -
Chris Cannam -
J M Needham -
Sebastian Tschöpel