Re: Speech shaped noise (Pierre Divenyi )

Subject: Re: Speech shaped noise
From:    Pierre Divenyi  <pdivenyi(at)EBIRE.ORG>
Date:    Sun, 20 Feb 2005 15:07:15 -0800

Hi Richard,

The problem with Bernhard's routine is that it **assumes** to know the
speech spectrum. It is easy to write Matlab code that generates a speech
spectrum noise using the canned fftfilt script. If it is placed in an
overlap-add loop, you can generate a speech noise file of arbitrary length
(minutes, if you so wish) by filtering a white noise vector (or file) with
the spectrum of any speech sample of virtually any length.

Pierre

Here it is:

At 10:54 AM 2/20/2005, Bernhard Seeber wrote:
>Hi Richard,
>
>here is a matlab-program to make CCITT-speech shaped, but not
>speech-modulated noise from white noise. A sample of Fastl noise
>(noise-like in long term spectrum and temporal envelope) can be
>downloaded from:
>http://www.mmk.ei.tum.de/persons/fas.html
>
>Bernhard
>
>
>
>>Date:    Sat, 19 Feb 2005 21:30:59 -0000
>>From:    "Richard H." <auditory(at)AUGMENTICS.COM>
>>Subject: Speech shaped noise .WAV or other sample?
>>
>>Hi,
>>
>>I am trying to test some modern digital hearing aids ... but they keep
>>(sensibly) identifying my puretones as feedback or noise, and
>>my broadband noise as junk too.
>>
>>I really need to use speech flavoured noise to prevent the anti-feedback
>>and anti-noise algorithms from activating.
>>
>>So ... does anyone know where I can find a 30-second or longer chunk of
>>speech-shaped noise to do my testing?
>>
>>Thanks!
>>
>>Richard (UK)
>
>
>--
>Dr.-Ing. Bernhard U. Seeber
>Department of Psychology
>University of California, Berkeley
>3210 Tolman Hall #1650
>Berkeley, CA 94720-1650, USA
>ph.: +1 510-643-8408
>fax: +1 510-642-5293
>http://www.bseeber.de
>
>
>function y = ccitt_filter(sig)
>% function y = ccitt_filter(sig)
>%
>% CCITT (ITU) standard G.227 defines a 'conventional
>% telephone signal' which equals the long-term spectrum
>% of speech. The standard defines a filter to simulate
>% the speech spectrum which is applied to the input signal
>% sig.
>%
>% sig    Input signal sig at 44100 Hz sampling rate
>% y      CCITT filtered signal sig
>%
>% Bernhard Seeber, 2005
>
>fs = 44100;
>
>%Generate filter coefficients
>%in S/Omega-plane
>%num_paper = [11638, 54050, 91238, 67280, 18400];
>%denom_paper = [1, 130, 4001, 36040, 400];
>
>%make inverted freq-response and shift to 0dB (at) 600Hz
>%num_coeff = ((2*pi*1000).^(-[4:-1:0])).*denom_paper;
>%denom_coeff = ((2*pi*1000).^(-[4:-1:0])).*num_paper;
>
>%for comparison, converted to Omega
>%num_coeff = [7.46722e-12, 2.17899e-7, 2.311086e-3, 10.7079, 18400];
>%denom_coeff = [6.41624e-16, 5.24087e-10, 1.01347e-4, 5.7359, 400];
>
>%scale coefficients
>%num_coeff = num_coeff / denom_coeff(1) * 10^(3.25/20);
>%denom_coeff = denom_coeff / denom_coeff(1);
>
>%freqs(num_coeff, denom_coeff);
>
>%convert to z-plane
>%[num_z, denom_z] = bilinear(num_coeff, denom_coeff, fs);
>
>%freqz(num_z, denom_z, 1024, fs);
>
>num_z =
>[0.00396790391508   0.00032556793042  -0.00314367152058  -0.00104604251859
>-0.00008875919940];
>denom_z =
>[1.00000000000000  -3.39268359295324   4.31295903323020  -2.43473845585969
>  0.51493759484342];
>
>y = filter(num_z, denom_z, sig);