gusucode.com > 语音分析源码程序 > 语音分析源码程序/HTK MFCC MATLAB/mfcc/vec2frames.m
function [ frames, indexes ] = vec2frames( vec, Nw, Ns, direction, window, padding )
% VEC2FRAMES Splits signal into overlapped frames using indexing.
%
% B=vec2frames(A,M,N) creates a matrix B whose columns consist of
% segments of length M, taken at every N samples along input vector A.
%
% [B,R]=vec2frames(A,M,N,D,W,P) creates a matrix B whose columns
% or rows, as specified by D, consist of segments of length M, taken
% at every N samples along the input vector A and windowed using the
% analysis window specified by W. The division of A into frames is
% achieved using indexes returned in R as follows: B=A(R);
%
% Summary
%
% A is an input vector
%
% M is a frame length (in samples)
%
% N is a frame shift (in samples)
%
% D specifies if the frames in B are rows or columns,
% i.e., D = 'rows' or 'cols', respectively
%
% W is an optional analysis window function to be applied to
% each frame, given as a function handle, e.g., W = @hanning
% or as a vector of window samples, e.g., W = hanning( M )
%
% P specifies if last frame should be padded to full length,
% or simply discarded, i.e., P = true or false, respectively
%
% B is the output matrix of frames
%
% R is a matrix of indexes used for framing, such that division
% of A into frames is achieved as follows: B=A(R);
%
% Examples
%
% % divide the input vector into seven-sample-long frames with a shift
% % of three samples and return frames as columns of the output matrix
% % (note that the last sample of the input vector is discarded)
% vec2frames( [1:20], 7, 3 )
%
% % divide the input vector into seven-sample-long frames with a shift
% % of three samples and return frames as rows of the output matrix
% % (note that the last sample of the input vector is discarded)
% vec2frames( [1:20], 7, 3, 'rows' )
%
% % divide the input vector into seven-sample-long frames with a shift
% % of three samples, pad the last frame with zeros so that no samples
% % are discarded and return frames as rows of the output matrix
% vec2frames( [1:20], 7, 3, 'rows', [], true )
%
% % divide the input vector into seven-sample-long frames with a shift
% % of three samples, pad the last frame with white Gaussian noise
% % of variance (1E-5)^2 so that no samples are discarded and
% % return frames as rows of the output matrix
% vec2frames( [1:20], 7, 3, 'rows', false, { 'noise', 1E-5 } )
%
% % divide the input vector into seven-sample-long frames with a shift
% % of three samples, pad the last frame with zeros so that no samples
% % are discarded, apply the Hanning analysis window to each frame and
% % return frames as columns of the output matrix
% vec2frames( [1:20], 7, 3, 'cols', @hanning, 0 )
%
% See also FRAMES2VEC, DEMO
% Author: Kamil Wojcicki, UTD, July 2011
% usage information
usage = 'usage: [ frames, indexes ] = vec2frames( vector, frame_length, frame_shift, direction, window, padding );';
% default settings
switch( nargin )
case { 0, 1, 2 }, error( usage );
case 3, padding=false; window=false; direction='cols';
case 4, padding=false; window=false;
case 5, padding=false;
end
% input validation
if( isempty(vec) || isempty(Nw) || isempty(Ns) ), error( usage ); end;
if( min(size(vec))~=1 ), error( usage ); end;
if( Nw==0 || Ns==0 ), error( usage ); end;
vec = vec(:); % ensure column vector
L = length( vec ); % length of the input vector
M = floor((L-Nw)/Ns+1); % number of frames
% perform signal padding to enable exact division of signal samples into frames
% (note that if padding is disabled, some samples may be discarded)
if( ~isempty(padding) )
% figure out if the input vector can be divided into frames exactly
E = (L-((M-1)*Ns+Nw));
% see if padding is actually needed
if( E>0 )
% how much padding will be needed to complete the last frame?
P = Nw-E;
% pad with zeros
if( islogical(padding) && padding )
vec = [ vec; zeros(P,1) ];
% pad with a specific numeric constant
elseif( isnumeric(padding) && length(padding)==1 )
vec = [ vec; padding*ones(P,1) ];
% pad with a low variance white Gaussian noise
elseif( isstr(padding) && strcmp(padding,'noise') )
vec = [ vec; 1E-6*randn(P,1) ];
% pad with a specific variance white Gaussian noise
elseif( iscell(padding) && strcmp(padding{1},'noise') )
if( length(padding)>1 ), scale = padding{2};
else, scale = 1E-6; end;
vec = [ vec; scale*randn(P,1) ];
% if not padding required, decrement frame count
% (not a very elegant solution)
else
M = M-1;
end
% increment the frame count
M = M+1;
end
end
% compute index matrix
switch( direction )
case 'rows' % for frames as rows
indf = Ns*[ 0:(M-1) ].'; % indexes for frames
inds = [ 1:Nw ]; % indexes for samples
indexes = indf(:,ones(1,Nw)) + inds(ones(M,1),:); % combined framing indexes
case 'cols' % for frames as columns
indf = Ns*[ 0:(M-1) ]; % indexes for frames
inds = [ 1:Nw ].'; % indexes for samples
indexes = indf(ones(Nw,1),:) + inds(:,ones(1,M)); % combined framing indexes
otherwise
error( sprintf('Direction: %s not supported!\n', direction) );
end
% divide the input signal into frames using indexing
frames = vec( indexes );
% return if custom analysis windowing was not requested
if( isempty(window) || ( islogical(window) && ~window ) ), return; end;
% if analysis window function handle was specified, generate window samples
if( isa(window,'function_handle') )
window = window( Nw );
end
% make sure analysis window is numeric and of correct length, otherwise return
if( isnumeric(window) && length(window)==Nw )
% apply analysis windowing beyond the implicit rectangular window function
switch( direction )
case 'rows', frames = frames * diag( window );
case 'cols', frames = diag( window ) * frames;
end
end
% EOF