Cascade Autocorrelation Model of Pitch Perception

From Emcap


Cascade Autocorrelation Model of Pitch

by Emili Balaguer-Ballester, Susan Denham, Ray Meddis

University of Plymouth, University of Essex



What is the Module?

It is an extension of the autocorrelation models of pitch perception which explains a wide range of perceptual experiments, not accounted for in previous approaches. It uses an aditional, longer temporal scale and a new auditory peripheral model.

What are the inputs, outputs and parameters?

Is a suite of matlab functions with different inputs-outputs. Please see 'How to use' section for a comprehensive description.

System requirements

Matlab, any recent version, and signal processing toolbox.

How to Cite

'A Cascade Autocorrelation Method of Pitch Perception", Journal of the Acoustic Society of America, to appear in October 2008.

The original publication is on-line available in

How to use it

Last update on August 3, 2008. Set of functions for testing several pitch perception experiments. Will be frequently updated. Please let me know if any problems. See updates in my web page or 'google' my name (Emili Balaguer-Ballester) if the page does not exists, or contact me by email.

Work made by Ray Meddis, Susan Denham and myself. See more details in corresponding paper.

After download and unzip this content from

,then you browse to '...\MAP1_6_Emili\private workspaces\EmiliWorkSpace' , which contains the functions I have prepared.


tests any alternating click train at a given peak level. There are various examples in this folder at different levels, e.g.


gives the results for a 4-6 alternating click, filtered and with pink noise added at a global level of 78 dB rms,


the same but at a level of 65 dB rms, etc. Apart form the pitch prediction using the peak (which is typically the more reliable pitch predictor for irregular and noisy stimuli) and the MSE method, there is also an analysis of the CAP responses using a method developed by Robert Ferry& Ray Meddis (JASA, 2007).

The function


generates many different click train realisations and shows some statistics. This function can be modified to test any click train. If it is not clear how to do that in the code, you could tell me, if you want to, which conditions you want to test and I will send you back this same function using the most suitable parameters.

The next functions:




permit to reproduce Oxenham et al 2004 stimuli/responses for any conditions (any carrier frequency/es and modulation frequency/es, noise level, etc.), e.g.

>>CarrierFrequencies=[4000];%Set three values for the multi tone case >>ModulationFrequencies=[100];%Set three values for the multi tone case >>Level=65;%Peak level >>NoiseLevel=38;%rms level >>TestTransposedSingleTone(CarrierFrequencies,ModulationFrequencies,Level,NoiseLevel);%Use TestTransposedMultiTone.m for the multi-tone case

Finally, the function


admits any stimuli, but do not uses the MSE criterion as there are no templates valid for any arbitrary stimulus. It is not thoroughly tested, I did it quickly in case you need it. Please let me know if any problems or if you want to adjust some of the functions. Thanks very much,

Emili Balaguer-Ballester