[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

AW: [AUDITORY] need roughness calculation results from different systems



Dear Densil, 

as I already said, Aures and D&W algorithm is essentilly the same. As D&W heavily rely on Aures's model they have left out implementation details that were unchanged. We were able to implement a running version of the Aures algorithm. Both articles combined should be sufficient to write an implementation except for the sad fact that you would have to estimate the filter curves from printed diagrams. 

I believe that roughness is caused by physical interaction of travelling waves on the basilar membrane. Thus models based on the analysis of isolated sine waves can only model a simplified subset of sounds where the interaction of the sines is linear or non existent. But model output should be at least comparable for the sine complexes the dissonance models were designed for. 

Best regards
Hannes


----- Ursprüngliche Mail ---- 
Von: Densil Cabrera <d.cabrera@xxxxxxxxxxxxxxxx> 
An: Hannes Löschke <hannes_loeschke@xxxxxxxx> 
CC: Sam Ferguson <sferguson@xxxxxxxxxxxxxxxx> 
Gesendet: Mittwoch, den 25. Juli 2007, 23:19:16 Uhr 
Betreff: FW: [AUDITORY] need roughness calculation results from different systems 


Dear Hannes, 

We would be interested if you reach any conclusions re PsySound3 implementation of D&W roughness because it is still really a beta version at present. The code was supplied by Dik Hermes, and adapted for the platform (after we tried to implement D&W?s algorithm locally, but found some intractable problems due to what we think is information missing from D&W?s article). 

By the way, PsySound2 implemented an adaptation of Hutchinson & Knopoff?s ?dissonance? model, as well as Sethares?. While acoustic dissonance should be a subset of roughness, our early tests suggest that there is not much of a relationship between these models and D&W roughness in terms of output. Of course, these models were designed primarily for theoretical analysis, rather than analysis of real sound waves, so the spectral analysis step probably has a big and possibly unanticipated effect on the results. I used the peak extraction algorithm proposed by Terhardt et al in their pitch model for these dissonance models, as well as just the raw FFT values. 

By the way, our aim for PsySound3 will be to have as many models as we can easily implement, all running in the same program. However, currently our programming effort is concentrated in other areas (modifying the basic program infrastructure, rather than focusing on analysis models). 

Best regards 
Densil 

_____________________________________ 
Densil Cabrera 
Head, Acoustics Research Laboratory 
Associate Dean (Research) 
Faculty of Architecture, Design and Planning 
University of Sydney NSW 2006 Australia 
Tel. +612 9351 5267 Skype densilcabrera 
Fax. +612 9351 3031 densil@xxxxxxxxxxx 
www.arch.usyd.edu.au/~densil/ 



From: AUDITORY - Research in Auditory Perception [mailto:AUDITORY@xxxxxxxxxxxxxxx] On Behalf Of Hannes Löschke 
Sent: Wednesday, 25 July 2007 9:03 PM 
To: AUDITORY@xxxxxxxxxxxxxxx 
Subject: AW: [AUDITORY] need roughness calculation results from different systems 

Aures is the basis for Daniel&Weber. Both algorithms are described in detail in Acustica/acta acustica. The references are 
Aures, W.: Ein Berechnungsverfahren der Rauhigkeit. Acustica 58, pp. 268-281 
Daniel, P.; Weber, R.: Psychoacoustical Roughness: Implementation of an Optimized Model. Acustica 83, pp. 113-123 

Both models are based on analysis of the modulation index in critical bandwith channels. Phase effects are taken into account by correlation between the channel signals. Both models are suffering from miscalculations for weak signals but are to my knowlegde the only algorithms outh there that are described in pretty much every detail of implementation. Apart from that, most comercially avaliable systems have roughness calculation implemented based on Aures' algorithm. So this calculation should have provided the best comparability between results of different systems. 
This is where "correct implementation" comes into play. Aures and Daniel&Weber provided a lot of detail, which pedicts most of the model behaviour. Obviously there are differences in calculation output for the different model implementations. Also there are known differences in various implementation details, but due to similarity in model structure we believe that there should be a greater similarity between the results. The question that arises from that is: What ist the implementation, that correctly calculates what the authors intended. 

Our proprietary model was developed to overcome the known problems of Aures and is conceptually pretty similar to Leman's SIM (which we didn't know of while developing the model). We're currently conducting experiments to support and tune our model. 

The CF90 is a workstation for psychoacoustic measurements from the early 90's manufactured by Cortex Instruments. Cortex was bought by 01dB. 

----- Ursprüngliche Mail ---- 
Von: PORRES <mentalosmosis@xxxxxxxxx> 
An: Hannes Löschke <hannes_loeschke@xxxxxxxx> 
Gesendet: Dienstag, den 24. Juli 2007, 19:05:39 Uhr 
Betreff: Re: AW: AW: AW: [AUDITORY] need roughness calculation results from different systems 

My model seems to go for another direction... it is based on plomp & Levelt's Results, Like Kameoka & Kurigawa, Like Hutchinson & Knopof, Like Sethares', Like Vassilakis', Like Barlow's... and some others maybe... 

Wel... I haven't been able to study Aures' algorith yet, could you direct me to his documentations somewhere? 

I know that like his... there are the work of Daniel & Weber! Which calculate roughness as a sum of the energy of the beating frequencies in auditory channels. But also similarly... Pressnitzer, . Pressnitzer uses a similar model but takes into account the effects of phase modulation in different auditory channels. Regarding phase and auitory channels, Leman uses a synchronization index model of sensory dissonance and roughness that accounts for the degree of phase-locking to a particular frequency that is present in the neural patterns. 

The procedure of suming roughness values fro mauditory channels seem to be the same procedure after all for the ones based on Plomp & Levelt's result, because they define this curve of roughness value for a single auditory channel... 

As I said, I wasn't able to study Aures and all the others that are similar, but I would like to check it out and compare the methodologies to mine. 

Pressnitzer offers his model in puredata, check it at http://cognition.ens.fr/Audition/tools/realtime/ 

So, tell me sonething, what is the "Cortex CF90"? What do you mean by "correct implementation"? What is the model to be correctly implemented? what is your second step based at?And again... please direct me to the information of Aures' and the similar approaches that you are working on... 

Cheers 
Alex 

Hannes Löschke <hannes_loeschke@xxxxxxxx> wrote: 
Our first step was to implement the pretty well documented Aures algorithm. Verfication against our Cortex CF90 revealed remakable differences, which lead to my request here on the list. Basically I'm now trying to figure out what the correct implementation is. Our results are pretty similar to the graphs calculated by the daniel&weber implementation found in PsySound3. Are there any other known implementations of this algorithm? 

The second step will be the implementation of a second approach that is currently justified by hearing tests. 
----- Ursprüngliche Mail ---- 
Von: PORRES <mentalosmosis@xxxxxxxxx> 
An: Hannes Löschke <hannes_loeschke@xxxxxxxx> 
Gesendet: Freitag, den 20. Juli 2007, 17:07:12 Uhr 
Betreff: Re: AW: AW: [AUDITORY] need roughness calculation results from different systems 
ok, so lets check each other's implementation against each other :) 

I will check the file, I am curious on whar kind and method you are adopting. 

cheers 
alex 

Hannes Löschke <hannes_loeschke@xxxxxxxx> wrote: 
Hi Alex, 

I'm sorry I missed the last bit of your message. I'm currently checking our model implementation against other implementations and we're still compiling the data of our listening tests. So there isn't really any data to publish at this point. 

The file can be downloaded at http://www.yousendit.com/download/OGhma3NiTEQ1aVkwTVE9PQ. 

Thanks for your support 
Hannes 
----- Ursprüngliche Mail ---- 
Von: PORRES <mentalosmosis@xxxxxxxxx> 
An: Hannes Löschke <hannes_loeschke@xxxxxxxx>; AUDITORY@xxxxxxxxxxxxxxx 
Gesendet: Donnerstag, den 19. Juli 2007, 19:57:49 Uhr 
Betreff: Re: AW: [AUDITORY] need roughness calculation results from different systems 

hi hans, I dont know why I havent seen anything about this test file of yours... 

I wrote to the list telling about my roughness model in puredata, did you read it? Anyway, send me this file of yours, I would like to test it on my system 

cheers 

alex 

Hannes Löschke <hannes_loeschke@xxxxxxxx> wrote: 
Thank you all for giving me pointers to freely available packages and performing calculation on my test file. I still like to see some more output of commercial packages that are not available to everyone. 

btw.: Yesterday I had a message on the topic being blocked by the spam filter. So everyone that sent me a message I haven't answered, please send that again. 

Thank you all. 
Hannes 
----- Ursprüngliche Mail ---- 
Von: Hannes Löschke <hannes_loeschke@xxxxxxxx> 
An: AUDITORY@xxxxxxxxxxxxxxx 
Gesendet: Dienstag, den 17. Juli 2007, 11:00:22 Uhr 
Betreff: [AUDITORY] need roughness calculation results from different systems 
Dear list, 

we're currently researching on algorithms for roughness calculation. There are various models out there with different problems and behaviour. As our ressources are limited we can't gain access to all possible systems, so I'm asking you to help me out. 
I need the roughness vs. time and specific roughness vs. time plots from as many different systems as possible. I have prepared a test file and will send it to everyone willing to help me out. The file is 10sek. long, sampled at 48kHz/16bit/mono. 

Thank you 

Hannes Löschke 

Institut für Musikinstrumentenbau 
Klingenthaler Straße 42 
08267 Zwota 
http://www.ifm-zwota.de 




Was ist Glück? Schlafen Fische überhaupt? Die Antworten gibt?s auf Yahoo! Clever. 





Die etwas anderen Infos rund um das Thema Reisen. BE A BETTER WELTENBUMMLER! 




Choose the right car based on your needs. Check out Yahoo! Autos new Car Finder tool. 





Heute schon einen Blick in die Zukunft von E-Mails wagen? Versuchen Sie´s mit dem neuen Yahoo! Mail. 




Got a little couch potato? 
Check out fun summer activities for kids. 





Beginnen Sie den Tag mit den neuesten Nachrichten. Machen Sie Yahoo! zu Ihrer Startseite! 




Ready for the edge of your seat? Check out tonight's top picks on Yahoo! TV. 





Heute schon einen Blick in die Zukunft von E-Mails wagen? Versuchen Sie´s mit dem neuen Yahoo! Mail. 





Heute schon einen Blick in die Zukunft von E-Mails wagen? Versuchen Sie´s mit dem neuen Yahoo! Mail.


      __________________________________
Machen Sie Yahoo! zu Ihrer Startseite. Los geht's: 
http://de.yahoo.com/set


      __________________________________  Die etwas anderen Infos rund um das Thema Reisen. BE A BETTER WELTENBUMMLER!  www.yahoo.de/clever