Current time-stretch algos applied to impulse responses ?
- KVRAF
- Topic Starter
- 6244 posts since 25 May, 2002 from Bobo-dioulasso\BF__Geneva/CH
Let me apologise first because i already opened a very similar topic at least once in this forum
But i was thinking more closely on the problem and one point that appear to my mind was : what might currently be the best time-stretch engine that doesn't create unwanted artifacts while applied to such spaces ?
Questions of psychoacoustic consideration : specialised time-stretch algorithm suitable for these peculiar types of signal/sounsource...different cases if you want to independently modify monophonic signals or polyphonic instrument signals for instances !
But i was thinking more closely on the problem and one point that appear to my mind was : what might currently be the best time-stretch engine that doesn't create unwanted artifacts while applied to such spaces ?
Questions of psychoacoustic consideration : specialised time-stretch algorithm suitable for these peculiar types of signal/sounsource...different cases if you want to independently modify monophonic signals or polyphonic instrument signals for instances !
-
- KVRAF
- 2462 posts since 15 Apr, 2004 from Capital City, UK
This is a slow topic..
But I've also considered this over the years, and to my knowledge, there are no time stretching algorithms which do not produce artifacts or uncontrollable phase-shifts in some or all frequencies allowing for IR modifications, unless one seeks the artifacts themselves! Few enough artifacts to not affect the source in unwanted ways can, I doubt, be achieved. If it were I suspect one or more of the major impulse response players (AudioEase, probably Waves) would have this as a feature.
At most, and Space Designer offers this, is a hard ratio driven multiplication of the file; double, or half, or quarter, which obviously results in the impulse changing character and tone. No phase issues will occur this way, as you probably know.
So yeah, no, one can't stretch/shrink impulse responses succesfully at the moment. It may be that IRCAM or some clever boffins at a DSP research institute somewhere have developed the perfect time-stretching algo suitable for this kind of operation and have yet to work out how to monetise it!
Although I am ALWAYS willing to be corrected
But I've also considered this over the years, and to my knowledge, there are no time stretching algorithms which do not produce artifacts or uncontrollable phase-shifts in some or all frequencies allowing for IR modifications, unless one seeks the artifacts themselves! Few enough artifacts to not affect the source in unwanted ways can, I doubt, be achieved. If it were I suspect one or more of the major impulse response players (AudioEase, probably Waves) would have this as a feature.
At most, and Space Designer offers this, is a hard ratio driven multiplication of the file; double, or half, or quarter, which obviously results in the impulse changing character and tone. No phase issues will occur this way, as you probably know.
So yeah, no, one can't stretch/shrink impulse responses succesfully at the moment. It may be that IRCAM or some clever boffins at a DSP research institute somewhere have developed the perfect time-stretching algo suitable for this kind of operation and have yet to work out how to monetise it!
Although I am ALWAYS willing to be corrected
- KVRAF
- Topic Starter
- 6244 posts since 25 May, 2002 from Bobo-dioulasso\BF__Geneva/CH
-
Sampleconstruct Sampleconstruct https://www.kvraudio.com/forum/memberlist.php?mode=viewprofile&u=191286
- KVRAF
- 16138 posts since 12 Oct, 2008 from Here and there
-
- KVRAF
- 2462 posts since 15 Apr, 2004 from Capital City, UK
-
Sampleconstruct Sampleconstruct https://www.kvraudio.com/forum/memberlist.php?mode=viewprofile&u=191286
- KVRAF
- 16138 posts since 12 Oct, 2008 from Here and there
The times I used the stretch function, it always produced satisfactory/good results.CinningBao wrote:No probs Krakatau Gotta share what we learn..
Simon, I'm not a Melda user so I didn't realise this.
Would you say it is suitable for the expected detail of convolution reverbs?
-
- KVRAF
- 2462 posts since 15 Apr, 2004 from Capital City, UK
Interesting - the manual says:
but also
So my guess is that it's just slowing down/speeding up the IR like a tape/record - it doesn't look like there are any complex time-stretching algos present. Since, though, most reverb IRs just sound like noise, speedingup/slowingdown a little bit wouldn't affect the result too much, but extreme settings will be fun and experimental!
Code: Select all
Stretch
Stretch resizes the impulse response. As a result you may observe not only a change of length, but also a change of pitch.
Code: Select all
Override length
Override length changes the length of the IR to the specified length. It can be used for various creative effects, but the most typical use is to synchronize an IR consisting of a loop to the current tempo. It is ignored when the length is minimum, hence 0ms, or OFF. Please note that since its operation is the same as the Pitch shift parameter, it is strongly advised not to use both parameters at once.
Pitch shift
Pitch shift changes the length of the IR and as a result causes a pitch shift specified in semitones. Please note that since its operation is the same as Override length parameter, it is strongly adviced not to use both parameters at once.
- KVRian
- 698 posts since 7 Dec, 2009 from GWB
There are folks using PaulStretch on IRs.
- KVRAF
- 3426 posts since 15 Nov, 2006 from Pacific NW
I would stick with "repitching" the impulse (i.e. interpolation so that it plays back at a slower or faster rate), possibly combined with some filtering and/or nonlinear harmonic enhancement to recreate the original frequency response.
Most time-stretching algorithms, that seek to preserve the pitch of a signal while changing the duration, make the presumption that the signal is largely harmonic. Impulses are usually inharmonic. So the types of tricks used to time-stretch harmonic signals (i.e. tracking the period of the signal to determine best splicing point, resynthesizing the signal using a sum of sinusoids) would fail on inharmonic signals.
IIRC, Jonathan Abel (of Universal Audio fame, currently a researcher at CCRMA) has done some recent work on extending impulse responses by calculating the frequency response in multiple bands, and using Gaussian noise to resynthesize the tail. This approach would probably be the best for standard reverberation purposes, but I don't know if it has made it out of the research stage.
Sean Costello
Most time-stretching algorithms, that seek to preserve the pitch of a signal while changing the duration, make the presumption that the signal is largely harmonic. Impulses are usually inharmonic. So the types of tricks used to time-stretch harmonic signals (i.e. tracking the period of the signal to determine best splicing point, resynthesizing the signal using a sum of sinusoids) would fail on inharmonic signals.
IIRC, Jonathan Abel (of Universal Audio fame, currently a researcher at CCRMA) has done some recent work on extending impulse responses by calculating the frequency response in multiple bands, and using Gaussian noise to resynthesize the tail. This approach would probably be the best for standard reverberation purposes, but I don't know if it has made it out of the research stage.
Sean Costello
- KVRAF
- Topic Starter
- 6244 posts since 25 May, 2002 from Bobo-dioulasso\BF__Geneva/CH
exactly...this is the crucial pointvalhallasound wrote:
Most time-stretching algorithms, that seek to preserve the pitch of a signal while changing the duration, make the presumption that the signal is largely harmonic. Impulses are usually inharmonic. So the types of tricks used to time-stretch harmonic signals (i.e. tracking the period of the signal to determine best splicing point, resynthesizing the signal using a sum of sinusoids) would fail on inharmonic signals.
Thank you Sean for all the relevant infos, instructive and much appreciatedvalhallasound wrote: IIRC, Jonathan Abel (of Universal Audio fame, currently a researcher at CCRMA) has done some recent work on extending impulse responses by calculating the frequency response in multiple bands, and using Gaussian noise to resynthesize the tail. This approach would probably be the best for standard reverberation purposes, but I don't know if it has made it out of the research stage.
Sean Costello
- KVRAF
- 3426 posts since 15 Nov, 2006 from Pacific NW
My pleasure! Here's some slides from Nicholas Bryan and Jonathan Abel on extending impulse responses:Krakatau wrote:Thank you Sean for all the relevant infos, instructive and much appreciatedvalhallasound wrote: IIRC, Jonathan Abel (of Universal Audio fame, currently a researcher at CCRMA) has done some recent work on extending impulse responses by calculating the frequency response in multiple bands, and using Gaussian noise to resynthesize the tail. This approach would probably be the best for standard reverberation purposes, but I don't know if it has made it out of the research stage.
Sean Costello
https://ccrma.stanford.edu/~njb/researc ... Slides.pdf
- KVRAF
- Topic Starter
- 6244 posts since 25 May, 2002 from Bobo-dioulasso\BF__Geneva/CH
Thank you, seems very appropriate to enlight even relatively profane minds to the subject,valhallasound wrote: Here's some slides from Nicholas Bryan and Jonathan Abel on extending impulse responses:
https://ccrma.stanford.edu/~njb/researc ... Slides.pdf
i might find the time to read this carefully...excellent !
-
- KVRAF
- 2462 posts since 15 Apr, 2004 from Capital City, UK