THE AUDITORY MODELING TOOLBOX

Applies to version: 1.1.0

View the code

Go to function

HAUTH2020 - Blind equalization cancellation model

Usage

out_struct = hauth2020('RequiredSignal',mixed_input,'OptionalSignal',...[OptionalSignal1 OptionalSignal2]...,OptionalSignalN,'model_params',model_params)

Input parameters

MixedSig two-channel matrix containing signal + noise
fs the sampling frequency of RequiredSignal [Hz]

Output parameters

out_struct Structure containing the processed signals, in the order [Sel Min Max]

Description

Additional input parameters:

'OptSigs= OptSig matrix containing one or more optional two channel signals'

'fmin= fmin center frequency of the lowest filter of the gammatone filterbank [Hz]'

'fmax= fmax center frequency of the highest filter of the gammatone filterbank [Hz]'

'f_target' , f_target specified frequency that will have a matched gammatone filter [Hz]

'ERB_factor= ERB_factor gammatone filter bandwidth'

Hauth2020 accepts the following flags:

'binauralinaccuracies'
 enable binaural processing inaccuracies
'no_binauralinaccuracies'
 disable binaural processing inaccuracies
'longterm' SRMR-decision making considers whole signal
'shortterm' blockwise SRMR-decision making based on same time constants as EC and BE processing

Parameter description:

'RequiredSignal' is the mixture of speech and noise, which is a two channel (left,right) matrix (mixed_input)
'OptionalSignal' One or more optional two channel signals, which are processed in the same way as the 'RequiredSignal'. They have to be arranged in a single matrix such that always two columns correspond to the left and right ear signals [left1 right1 left2 right2 ... leftN rightN] (e.g. if you need clean speech and noise for your back-end)
'model_params'

Structure containing model parameters like frequency range etc.

out_struct Structure containing the processed signals, in the order [Sel Min Max]
e.g. out_struct.signals.Mixsigsynfin contains the EC/BE processed mixed signal, where the best strategy was chosen blindly

The SII used in the study by Hauth et al. (2020) requires band specific levels. If optional signals are present, the levels of all optional signals after EC processing and better ear selection are part of the output structure.

References:

C. F. Hauth, S. C. Berning, B. Kollmeier, and T. Brand. Modeling binaural unmasking of speech using a blind binaural processing stage. Trends in Hearing, 2020. [ http ]