Pyramind

Im looking for a software that can do "batch" analysis to huge number of short(1-3 sec) audio files based on:

* spectral content(pitched? if pitched what is the perceived frequency/pitch? is it unpitched? is it noisy?)

* Loudness, and maybe dynamic data(the difference between the peak point and noise floor)

and maybe other stuff but these are essential.

And it should be able to arrange these files into folders based on this analysis! Something like:

* files shorter than 2 seconds go to this folder

things like that... It does not have to do them all at the same time, I can run many iterations under sorted folders, like first seperating the ones shorter than say, 2 seconds then seperating unpitched ones from them...

I'm totally lost in this one. I'm pretty much skilled at max/msp, and Pure Data but I've never dealt with non realtime analysis and extensive file operations so I'm not really sure if these are the right tools to program something like that.

Is there any software that can do stuff like this?

Im looking for a software that can do "batch" analysis to huge number of short(1-3 sec) audio files based on: 

* spectral content(pitched? if pitched what is the perceived frequency/pitch? is it unpitched? is it noisy?)
* Loudness, and maybe dynamic data(the difference between the peak point and noise floor)
* Length (should be straightforward)

* files shorter than 2 seconds go to this folder
* unpitched sounds to this folder

which-software-can-do-this

You could use FTM and Gabor to create SDIF-like descriptions of the 

files, and then shell or applescript to sort depending on the results.

I'd suggest that you download and look into FTM and Gabor, and also 

download CataRT and study that to see how such analysis can be carried out:

Don't expect to mamage all of this in one day, though.

> Im looking for a software that can do "batch" analysis to huge number of short(1-3 sec) audio files based on: 

> * spectral content(pitched? if pitched what is the perceived frequency/pitch? is it unpitched? is it noisy?)

> * Loudness, and maybe dynamic data(the difference between the peak point and noise floor)

> and maybe other stuff but these are essential.

> And it should be able to arrange these files into folders based on this analysis! Something like:

> * files shorter than 2 seconds go to this folder

> things like that... It does not have to do them all at the same time, I can run many iterations under sorted folders, like first seperating the ones shorter than say, 2 seconds then seperating unpitched ones from them...

> I'm totally lost in this one. I'm pretty skilled at max/msp, and Pure Data but I've never dealt with non realtime analysis and extensive file operations so I'm not really sure if these are the right tools to program something like that.

> Is there any software that can do stuff like this?

You could use FTM and Gabor to create SDIF-like descriptions of the 
files, and then shell or applescript to sort depending on the results.

I'd suggest that you download and look into FTM and Gabor, and also 
download CataRT and study that to see how such analysis can be carried out:

http://ftm.ircam.fr
http://imtr.ircam.fr/index.php/CataRT

Batuhan wrote:
> Hi,
>
> Im looking for a software that can do "batch" analysis to huge number of short(1-3 sec) audio files based on: 
>
> * spectral content(pitched? if pitched what is the perceived frequency/pitch? is it unpitched? is it noisy?)
> * Loudness, and maybe dynamic data(the difference between the peak point and noise floor)
> * Length (should be straightforward)
>
> and maybe other stuff but these are essential.
>
> And it should be able to arrange these files into folders based on this analysis! Something like:
>
> * files shorter than 2 seconds go to this folder
> * unpitched sounds to this folder
>
> things like that... It does not have to do them all at the same time, I can run many iterations under sorted folders, like first seperating the ones shorter than say, 2 seconds then seperating unpitched ones from them...
>
> I'm totally lost in this one. I'm pretty skilled at max/msp, and Pure Data but I've never dealt with non realtime analysis and extensive file operations so I'm not really sure if these are the right tools to program something like that.
>
> Is there any software that can do stuff like this?
>
> Thnks for your help!
>


although it may not do exactly what you want. It has a *lot* of

> Im looking for a software that can do "batch" analysis to huge

> number of short(1-3 sec) audio files based on:

> * spectral content(pitched? if pitched what is the perceived

> frequency/pitch? is it unpitched? is it noisy?)

> * Loudness, and maybe dynamic data(the difference between the

> And it should be able to arrange these files into folders based

> things like that... It does not have to do them all at the same

> time, I can run many iterations under sorted folders, like first

> seperating the ones shorter than say, 2 seconds then seperating

> I'm totally lost in this one. I'm pretty skilled at max/msp, and

> Pure Data but I've never dealt with non realtime analysis and

> extensive file operations so I'm not really sure if these are the

> right tools to program something like that.

   http://labrosa.ee.columbia.edu/meapsoft/

although it may not do exactly what you want.  It has a *lot* of
feature analysis capabilities, though.

>
> Hi,
>
> Im looking for a software that can do "batch" analysis to huge
> number of short(1-3 sec) audio files based on:
>
> * spectral content(pitched? if pitched what is the perceived
> frequency/pitch? is it unpitched? is it noisy?)
> * Loudness, and maybe dynamic data(the difference between the
> peak point and noise floor)
> * Length (should be straightforward)
>
> and maybe other stuff but these are essential.
>
> And it should be able to arrange these files into folders based
> on this analysis! Something like:
>
> * files shorter than 2 seconds go to this folder
> * unpitched sounds to this folder
>
> things like that... It does not have to do them all at the same
> time, I can run many iterations under sorted folders, like first
> seperating the ones shorter than say, 2 seconds then seperating
> unpitched ones from them...
>
> I'm totally lost in this one. I'm pretty skilled at max/msp, and
> Pure Data but I've never dealt with non realtime analysis and
> extensive file operations so I'm not really sure if these are the
> right tools to program something like that.
>
> Is there any software that can do stuff like this?
>
> Thnks for your help!
>


Trond, it looks like a pain to do but I think extracting data manually and scripting them would be my last resort if I cannot find an easy solution. So thanks for the info.

Bram, MEAPsoft looks amazing, It's very close to what I'm trying to achieve. I want to have a database of little sound chunks(1-2 seconds each) sliced and extracted from various sources that are arranged by their spectral and dynamic properties. I'm talking about thousands of slices here.

Then I'll access them by an interface that I'll write which will be able to layer them, window them, play them back by any means necessary. The interface will be controlled by a score, as this is a piece I'm currently working on.

I just need to have those files in correct places.

I'll check steinberg wavelabs batch processing and scripting capabilities.

The worst thing is that I remember stumbling upon a software that specialises at this task. But there is no way for me to find it again as the moment is very faded in my memory.

I'm hoping that someone here knows what that software is actually. : )

So thanks again for the help. MEAPsoft made me feel more exciting about this upcoming concept, piece and instrument/tool that I'm gonna make.

I just need to have those files in correct places.
I'll check steinberg wavelabs batch processing and scripting capabilities.

> Bram, MEAPsoft looks amazing, It's very close to what I'm trying

> to achieve. I want to have a database of little sound chunks(1-2

> seconds each) sliced and extracted from various sources that are

> arranged by their spectral and dynamic properties. I'm talking

I think MEAPsoft is designed to work on single soundfiles, you might

want to consider combining all your slices into a single file, each

> Then I'll access them by an interface that I'll write which will

> be able to layer them, window them, play them back by any means

> necessary. The interface will be controlled by a score, as this

I believe that Jeff Snyder (jos2101 - at - columbia -dot- edu) and

Sam Pluta (spluta - at - gmail -dot- com) have worked on using the

MEAPsoft data in max/msp and SC3 environments. Might want to check

> So thanks again for the help. MEAPsoft made me feel more exciting

> about this upcoming concept, piece and instrument/tool that I'm

It's pretty impressive stuff. Dan Ellis and those guys actually

> Bram, MEAPsoft looks amazing, It's very close to what I'm trying
> to achieve. I want to have a database of little sound chunks(1-2
> seconds each) sliced and extracted from various sources that are
> arranged by their spectral and dynamic properties. I'm talking
> about thousands of slices here.

I think MEAPsoft is designed to work on single soundfiles, you might
want to consider combining all your slices into a single file, each
one at a set timepoint.

> Then I'll access them by an interface that I'll write which will
> be able to layer them, window them, play them back by any means
> necessary. The interface will be controlled by a score, as this
> is a piece I'm currently working on.

I believe that Jeff Snyder (jos2101 - at - columbia -dot- edu) and
Sam Pluta (spluta - at -  gmail -dot- com) have worked on using the
MEAPsoft data in max/msp and SC3 environments.  Might want to check
with them about their results.

> So thanks again for the help. MEAPsoft made me feel more exciting
> about this upcoming concept, piece and instrument/tool that I'm
> gonna make.

It's pretty impressive stuff.  Dan Ellis and those guys actually
know what they're doing.

There's a dedicated mailing list to FTM, and Norbert, Diemo and the rest 

> Trond, it looks like a pain to do but I think extracting data manually and scripting them would be my last resort if I cannot find an easy solution. So thanks for the info.

> Bram, MEAPsoft looks amazing, It's very close to what I'm trying to achieve. I want to have a database of little sound chunks(1-2 seconds each) sliced and extracted from various sources that are arranged by their spectral and dynamic properties. I'm talking about thousands of slices here.

> Then I'll access them by an interface that I'll write which will be able to layer them, window them, play them back by any means necessary. The interface will be controlled by a score, as this is a piece I'm currently working on.

> I just need to have those files in correct places.

> I'll check steinberg wavelabs batch processing and scripting capabilities.

> The worst thing is that I remember stumbling upon a software that specialises at this task. But there is no way for me to find it again as the moment is very faded in my memory.

> I'm hoping that someone here knows what that software is actually. : )

> So thanks again for the help. MEAPsoft made me feel more exciting about this upcoming concept, piece and instrument/tool that I'm gonna make.

There's a dedicated mailing list to FTM, and Norbert, Diemo and the rest 
are very helpful.

Batuhan wrote:
> Thanks for the responses!
>
> Trond, it looks like a pain to do but I think extracting data manually and scripting them would be my last resort if I cannot find an easy solution. So thanks for the info.
>
> Bram, MEAPsoft looks amazing, It's very close to what I'm trying to achieve. I want to have a database of little sound chunks(1-2 seconds each) sliced and extracted from various sources that are arranged by their spectral and dynamic properties. I'm talking about thousands of slices here.
>
> Then I'll access them by an interface that I'll write which will be able to layer them, window them, play them back by any means necessary. The interface will be controlled by a score, as this is a piece I'm currently working on.
>
> I just need to have those files in correct places.
> I'll check steinberg wavelabs batch processing and scripting capabilities.
>
> The worst thing is that I remember stumbling upon a software that specialises at this task. But there is no way for me to find it again as the moment is very faded in my memory.
>
> I'm hoping that someone here knows what that software is actually. : )
>
> So thanks again for the help. MEAPsoft made me feel more exciting about this upcoming concept, piece and instrument/tool that I'm gonna make.
>
> Batuhan
>


I think the biggest problem would be to establish if your file is pitched or

noise-based as this is a very relative continuum (noise-pitch). In a noisy

sample there may be hidden harmonic patterns and every pitched sound has

some noise components. It also depends on your definition of noisiness for

the present task. Perhaps LPC would be a good analysis model? I would be

very interested to know how this analysis can be achieved if someone has

> Im looking for a software that can do "batch" analysis to huge number of

> * Loudness, and maybe dynamic data(the difference between the peak point

> And it should be able to arrange these files into folders based on this

> things like that... It does not have to do them all at the same time, I

> can run many iterations under sorted folders, like first seperating the ones

> shorter than say, 2 seconds then seperating unpitched ones from them...

> I'm totally lost in this one. I'm pretty skilled at max/msp, and Pure Data

> but I've never dealt with non realtime analysis and extensive file

> operations so I'm not really sure if these are the right tools to program

I think the biggest problem would be to establish if your file is pitched or
noise-based as this is a very relative continuum (noise-pitch). In a noisy
sample there may be hidden harmonic patterns and every pitched sound has
some noise components. It also depends on your definition of noisiness for
the present task. Perhaps LPC would be a good analysis model? I would be
very interested to know how this analysis can be achieved if someone has
better knowledge than me?

On 30/07/07, Batuhan  wrote:
>
>
> Hi,
>
> Im looking for a software that can do "batch" analysis to huge number of
> short(1-3 sec) audio files based on:
>
> * spectral content(pitched? if pitched what is the perceived
> frequency/pitch? is it unpitched? is it noisy?)
> * Loudness, and maybe dynamic data(the difference between the peak point
> and noise floor)
> * Length (should be straightforward)
>
> and maybe other stuff but these are essential.
>
> And it should be able to arrange these files into folders based on this
> analysis! Something like:
>
> * files shorter than 2 seconds go to this folder
> * unpitched sounds to this folder
>
> things like that... It does not have to do them all at the same time, I
> can run many iterations under sorted folders, like first seperating the ones
> shorter than say, 2 seconds then seperating unpitched ones from them...
>
> I'm totally lost in this one. I'm pretty skilled at max/msp, and Pure Data
> but I've never dealt with non realtime analysis and extensive file
> operations so I'm not really sure if these are the right tools to program
> something like that.
>
> Is there any software that can do stuff like this?
>
> Thnks for your help!
>

I've mangled some music with MEAPsoft, the output data is in text format so that can be parsed, I'll see what I can do with that data, it looks pretty straightforward, but I'm not very experienced at parsing stuff. I may abandon slicing audios alltogether and build a system that reads from various places from tracks. Based on data from MEAPsoft.

Actually MEAPsoft does just what I need, but I need a custom (and possibly realtime) composer section that is all.

Peiman, the noise-pitch distinction that I'm talking about is not really about accuracy. Most data will not contain a single pitch, but there can be mathematical estimations that would be enough as I just want an "impression" of those concepts when discriminating between slices.

A slice from a string quartet and noise from a train will have different spectral characteristics for sure and it should be pretty easy to separate them if you look at the correct things.

Or a piano slice playing C and G with sustaining pitches low at background would not be identified as an f# easily(in average) so that kind of accuracy would be enough for my purposes.

well SDIF will be a DSP too when you think about it.

max has a nonrealtime mode or the option to upsample inside a poly and play the samples 10x faster than normal when building the database.

or you could even downsample the input /8 before you process.

if you downsample a buffer /8 and run it throuh an fffb filter with 50 bands on a 8x3 Ghz intel mac you should be able to process 50 files at a time in paralell.

myself i have never used the nonrealtime mode for such things, but its on my to do list. :)

Ok, I managed to work with meapsoft, and it works pretty well, altouhg I had to change the way I think about how this project will work.

Basically I use meapsoft to analyse and sort from low to high based on a selected criteria. It returns an .edl file which has [start time, segment length] that are sorted from low to high based on the criteria.

I import the file to MS Excel copy the necessary numbers to a text file from there. It's in this format:

They are sorted from low to high as I said. Then the rest is PD programming(as I work in linux these days).

Basically  I use meapsoft to analyse and sort from low to high based on a selected criteria. It returns an .edl file which has [start time, segment length] that are sorted from low to high based on the criteria.

"sample start time" "segment length"
"sample start time" "segment length"
"sample start time" "segment length"
...

They are sorted from low to high as I said. Then the rest is PD programming(as I work in linux these days).


Which software can do this?