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

stability in CMs

Dear List,

I'm working on analyses of confusion matrices measured twice from the same group of HI listeners 6 month interval.

I want to assess the stability of performance on both diagonals and off-diagonls as a function of SNR and of test-retest.

So far I found two relevant literatures (Dubno Dirks, 1982, Bilger and Wang, 1976), reporting correlations between test and retest.  

I wonder if any knows ways to show the stability other than correlations, and 

I wonder if anyone knows references other than those. 


Yang-soo Yoon
Ph.D. candidate
Auditory Perception at Dept. of SHS
Human Speech Recognition at Beckman Institute, UIUC
phone: 217-766-1367
--- Begin Message ---
Attached is some Matlab code written by Keansub Lee in my lab when we were
looking at making "personal audio lifelog" recordings unintelligible while
preserving the statistics needed for classifying environment and speakers.

It chops the sound into overlapping 200ms windows and permutes them
within a 1 second radius.  Basic usage is just y = scrambling(d,sr);
where d is the input audio and sr is the sampling rate.  See comments
in the file for changing the parameters (window length, overlap, scramble
radius etc.).

Also attached are a brief speech sound before and after scrambling.
Apologies to digest readers who will probably see a great deal of
meaningless ascii.

function y = scrambling(x, sr, h, w, wintype, s)

% x is a input signal and sr is samplingrate.  
% Using w-points[200ms] windows every h-points[100ms] over s-points[1 sec] radius, 
% scramble the input signal.  
% Make sure that the length of window should be smaller than 2*(frame's length) 
% For example, 
% [x,sr]= wavread('mdwh0_sx305.wav');
% h = sr*.1; 
% w = h*2;
% wintype = 'sinewin';
% s = sr;
% y = scrambling(x, sr, h, w, wintype, s);

% 2004-10-14 kslee@xxxxxxxxxxxxxxx

if nargin < 3
  h = sr*0.1; % 100ms, frame length
if nargin < 4
  w = 2*h; % 200ms, window length
if nargin < 5
  wintype = 'sinewin'; % window type
if nargin < 6
  s = sr; % 1s, segments length
switch wintype
case 'sinewin'
    wt = sin(pi*([0:(w-1)]' + 0.5)/w);;
case 'hanningwin'
    wt = hanning(w);

if (size(x,1) == 1)
  x = x';  % Convert X from row to column

%x= x(1:sr*3);
% the points of input signal
npts = length(x);

% the total number of segments within input signal
nsegs = floor(npts/s);

% the total number of frames within each segment
nhops = floor(s/h);

y = [];

for seg = 1:nsegs
    % Extract segment of signal
    xx = x((seg - 1)*s + [1:s]);
    % points to be scrambled and remained 
    xxs = xx(1:nhops*h,1);
    xxr = xx((nhops*h + 1):end,1);
    % Pad x with zeros to extract complete w-length windows from it
    xxs = [zeros((w-h)/2,1);xxs;zeros((w-h/2),1)];
    % save the scrambled signal
    yys = zeros((nhops+1)*h,1);
    % random permutation
    nrands = randperm(nhops);
    for hop = 1:nhops
        % Break the signal into w-length windows
        xxsw = xxs((nrands(hop) - 1)*h + [1:w]);
        % Apply window
        wxxsw = xxsw .* wt; 
        % reverse and save it
%        yys((hop - 1)*h + [1:w],1) = yys((hop - 1)*h + [1:w],1) + wxxsw([end:-1:1]);
        yys((hop - 1)*h + [1:w],1) = yys((hop - 1)*h + [1:w],1) + wxxsw;
    % trancate the # of zeros padding points to have a same size of original 
    yy = [yys(h/2+1:end-h/2,1);xxr];
    y = [y;yy];

Attachment: mpgr1_sx419.wav
Description: Wave audio

Attachment: scrambled.wav
Description: Wave audio

--- End Message ---