[microsound] audio analysis tools in max msp
mattyo
news at ostrowski.info
Thu May 5 13:25:10 EDT 2011
these are two separate problems you're thinking about: recognition and filtering. There are a number of tools in the FTM package that do fairly detailed sound analysis -- the entire catArt package is basically about doing analyses on grains -- and with enough patience, it's not impossible to identify particular sounds out of a field, particularly if it has some nice attacks or periodicity that the background doesn't.
that said, the difference between getting a computer to identify certain features that match a model and note that it's there is very different from actually extracting such a sound from a field. No one really understands how the cocktail party effect works, but it is a _very_ sophisticated function of our brains, An FFT is probably too blunt an instrument for the resynthesis of something like that, and my guess is that there are teams of geniuses funded by DARPA working on this for the surveillance industry -- i.e., nothing that's showing up in a max object anytime soon....
\M
On May 4, 2011, at 9:03 AM, Mark Pedersen wrote:
> hey eske,
>
> you can do a lot with Tristan Jehan's analyzer~ object - which is based in part on Miller Puckette's fiddle~ object
>
> it is not stereo, and it won't find a "sound-object" for you - but if you spend some time analysing a field recording and isolate the unique features of a sound - e.g. by looking beyond frequency filters to spectral features like noisiness, brightness and envelope features, you could then run two analyzer~ objects - one each for L & R - and then try to filter out whatever is not matching your target parameters and then measure the loudness in each channel. this is just a guess, and I'm sure there's much more knowledgeable folks who actually know what they're talking about.
>
> You may also want to look at IRCAM's gesture follower example patches that come as part of the free FTM package. Once example shows how to recognise voices using a Hidden Markov Model trained on MFCC's extracted from the sound samples. You could do the same thing to have a realtime system that can recognise the presence of a particular sound object. There's probably a fair bit of work to do in selecting audio features and training such a system.
>
> So, I reckon it's possible, but we're probably still waiting for someone to right [cocktailpartyeffect~] :-)
>
> regards,
>
> Mark
>
>
> On Wed, May 4, 2011 at 6:23 PM, eske nørholm <flemminglyst at gmail.com> wrote:
> hi microsounders,
>
> i'm interested in finding realtime audio analysis tools for realtime soundscape/soundwalk processing with max msp.
>
> how can we make maxmsp know where in a stereo field a "soundobject" comes from? how can we separate one "sound-object" from
> the noise, without it being to rude(simple highpass, lowpass filtering)?
>
> is there a [cocktailpartyeffect~] object or something alike out there...?
>
> realtime seems to be the key issue (in my mind)?
>
> any ideas, links, etc are very welcome!
>
> cheers,
> eske
>
> worgw.org
>
> _______________________________________________
> microsound mailing list
> microsound at microsound.org
> http://or8.net/mailman/listinfo/microsound
>
>
> _______________________________________________
> microsound mailing list
> microsound at microsound.org
> http://or8.net/mailman/listinfo/microsound
More information about the microsound
mailing list