THE AUDITORY MODELING TOOLBOX

Applies to version: 1.4.0

View the code

Go to function

LOCALIZATIONERROR - Compute psychoacoustic performance parameters for sound localization experiments

Usage:

[accL, precL, accP, precP, querr] = localizationerror(m)
[res, meta, par] = localizationerror(m,errorflag)

Input parameters:

m item list from a localization experiment.

Description:

localizationerror(m,errorflag) returns psychoacoustic performance parameters for experimental response patterns. m is a matrix. In each row of m, the information about the target and response in the columns must be provided.

[res, meta, par] = localizationerror(m,errorflag) calculates error metric res given by the string errorflag. meta contains additional metadata, par contains additional calculation results.

The columns of 'm' comprise:

  • 1:4 azi_target,ele_target,azi_response,ele_response
  • 5:8 lat_target,pol_target,lat_response,pol_response
  • 9 F/B-C resolved pol_response

The errorflag may be one of:

'accL' accuracy (average) in the lateral dimension
'precL' precision (std. dev.) in the lateral dimension
'precLcentral'
'precL' derived from responses close to horizontal meridian (+-30 deg elevation)
'accP' accuracy (average) in the polar dimension
'precP' precision (circular std. dev.) in the polar dimension
'querr' percentage of weighted polar errors > 45deg:[querr number_of_corrects total_number] = CalcAccPrec...
'accE' accuracy in the elevation
'absaccE' absolute accuracy in the elevation
'absaccL' absolute lateral accuracy
'accabsL' mean absolute lateral error
'accabsP' mean absolute polar error
'accPnoquerr' accP with quadrant errors removed
'precE' precision in the elevation
'querr90' precentage of weighted polar errors > 90deg
'precPmedian' unweighted polar precision error considering targets around the median plane only (+/-30deg lateral)
'precPmedianlocal' basing on precPmedian, quadrant errors (polar error >90 deg) are excluded. Similar to RMS local polar error (Middlebrooks, 1999)
'precPnoquerr' precP with quadrant errors removed
'rmsL' lateral RMS error according to Middlebrooks (1999)
'rmsPmedianlocal' unweighted polar RMS considering targets around the median plane only (+/-30deg lateral) and excluding quadrant errors (polar error > 90deg). Identical to RMS local polar error in Middlebrooks (1999).
'rmsPmedian' unweighted polar RMS considering targets around the median plane only (+/-30deg lateral). Includes quadrant errors.
'querrMiddlebrooks' quadrant errors as in Middlebrooks (1999). Use central positions within +/-30deg, querr is when polar error is >90deg [querr number_of_corrects total_number] = CalcAccPrec...
'corrcoefL' lateral correlation coeffficient. Output parameter: [cc, p]
'corrcoefP' polar correlation coeffficient, Only data within lateral +/-30deg are considered. Output parameter: [corr_coeff, p_of_significant_corr]
'SCC' spherical/spatial correlation coefficient (Carlile et al., 1997)
'gainLstats' performs linear regression to obtain lateral gain. See help of regress for a detailed description of the structure fields.
'gainL' lateral gain only
'pVeridicalL' Proportion of quasi-verdical responses in lateral dimension
'precLregress' lateral scatter around linear regression line (only quasi-veridical responses included).
'sirpMacpherson2000'
 performs an ad-hoc selective, iterative regression procedure (SIRP) as proposed in Macpherson & Middlebrooks (2000 JASA) in order to exclude outliers and reversals and isolate the main concentration of responses in the computation of the linear fits. As the original procedure does not necessarily converge, the implementation here runs up to 100 iterations and selects the iteration that includes the most responses or excludes the least outliers, Outlier distance criterion is 40deg, output parameters [f,r] correspond to regression results for frontal and rear hemisphere respectively, see help of regress for a detailed description of the structure fields.
'gainPfront' frontal polar gain only
'gainPrear' rear polar gain only
'gainP' polar gain averaged between front and back
'slopePfront' slope in degrees of regression line (frontal only)
'slopePrear' slope in degrees of regression line (rear only)
'slopeP' slope in degrees of regression line (front and back)
'pVeridicalPfront' proportion of quasi-verdical polar responses in the front
'pVeridicalPrear' proportion of quasi-verdical polar responses in the back
'pVeridicalP' proportion of quasi-verdical polar responses in total
'precPregressFront' polar scatter around linear regression line for the front (only quasi-veridical responses included)
'precPregressRear' same as precPregressFront but for the back
'precPregress' average between precPregressFront and precPregressRear
'perMacpherson2003' polar error rate used in Macpherson & Middlebrooks (2003). They measured the deviation of responses from the linear predictors obtained by an ad-hoc SIRP. Polar errors are defined by showing a deviation of >45deg (i.e., not being quasi-veridical) with respect to the linear flat stimulus prediction. Note that for this analysis the results from sirpMacpherson2000 are required and handled as localizationerror(m,f,r,'perMacpherson2003')

If no errorflag is provided, the function returns: accL, precL, precP, and querr

References:

J. C. Middlebrooks. Virtual localization improved by scaling nonindividualized external-ear transfer functions in frequency. J. Acoust. Soc. Am., 106:1493--1510, 1999.

E. A. Macpherson and J. C. Middlebrooks. Vertical-plane sound localization probed with ripple-spectrum noise. J. Acoust. Soc. Am., 114:430--445, 2003.

S. Carlile, P. Leong, and S. Hyams. The nature and distribution of errors in sound localization by human listeners. Hearing research, 114(1):179--196, 1997.