Mercurial > hg > camir-aes2014

annotate toolboxes/MIRtoolbox1.3.2/AuditoryToolbox/synlpc.m @ 0:e9a9cd732c1e tip

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
first hg version after svn
author wolffd
date Tue, 10 Feb 2015 15:05:51 +0000
parents
children
rev   line source
wolffd@0 1 function synWave = synlpc(aCoeff,source,sr,G,fr,fs,preemp)
wolffd@0 2 % USAGE: synWave = synlpc(aCoeff,source,sr,G,fr,fs,preemp);
wolffd@0 3 %
wolffd@0 4 % This function synthesizes a (speech) signal based on a LPC (linear-
wolffd@0 5 % predictive coding) model of the signal. The LPC coefficients are a
wolffd@0 6 % short-time measure of the speech signal which describe the signal as the
wolffd@0 7 % output of an all-pole filter. This all-pole filter provides a good
wolffd@0 8 % description of the speech articulators; thus LPC analysis is often used in
wolffd@0 9 % speech recognition and speech coding systems. The LPC analysis is done
wolffd@0 10 % using the proclpc routine. This routine can be used to verify that the
wolffd@0 11 % LPC analysis produces the correct answer, or as a synthesis stage after
wolffd@0 12 % first modifying the LPC model.
wolffd@0 13 %
wolffd@0 14 % The results of LPC analysis are a new representation of the signal
wolffd@0 15 % s(n) = G e(n) - sum from 1 to L a(i)s(n-i)
wolffd@0 16 % where s(n) is the original data. a(i) and e(n) are the outputs of the LPC
wolffd@0 17 % analysis with a(i) representing the LPC model. The e(n) term represents
wolffd@0 18 % either the speech source's excitation, or the residual: the details of the
wolffd@0 19 % signal that are not captured by the LPC coefficients. The G factor is a
wolffd@0 20 % gain term.
wolffd@0 21 %
wolffd@0 22 % LPC synthesis produces a monaural sound vector (synWave) which is
wolffd@0 23 % sampled at a sampling rate of "sr". The following parameters are mandatory
wolffd@0 24 % aCoeff - The LPC analysis results, a(i). One column of L+1 numbers for each
wolffd@0 25 % frame of data. The number of rows of aCoeff determines L.
wolffd@0 26 % source - The LPC residual, e(n). One column of sr*fs samples representing
wolffd@0 27 % the excitation or residual of the LPC filter.
wolffd@0 28 % G - The LPC gain for each frame.
wolffd@0 29 %
wolffd@0 30 % The following parameters are optional and default to the indicated values.
wolffd@0 31 % fr - Frame time increment, in ms. The LPC analysis is done starting every
wolffd@0 32 % fr ms in time. Defaults to 20ms (50 LPC vectors a second)
wolffd@0 33 % fs - Frame size in ms. The LPC analysis is done by windowing the speech
wolffd@0 34 % data with a rectangular window that is fs ms long. Defaults to 30ms
wolffd@0 35 % preemp - This variable is the epsilon in a digital one-zero filter which
wolffd@0 36 % serves to preemphasize the speech signal and compensate for the 6dB
wolffd@0 37 % per octave rolloff in the radiation function. Defaults to .9378.
wolffd@0 38 %
wolffd@0 39 % This code was graciously provided by:
wolffd@0 40 % Delores Etter (University of Colorado, Boulder) and
wolffd@0 41 % Professor Geoffrey Orsak (Southern Methodist University)
wolffd@0 42 % It was first published in
wolffd@0 43 % Orsak, G.C. et al. "Collaborative SP education using the Internet and
wolffd@0 44 % MATLAB" IEEE SIGNAL PROCESSING MAGAZINE Nov. 1995. vol.12, no.6, pp.
wolffd@0 45 % 23-32.
wolffd@0 46 % Modified and debugging plots added by Kate Nguyen and Malcolm Slaney
wolffd@0 47
wolffd@0 48 % (c) 1998 Interval Research Corporation
wolffd@0 49 % A more complete set of routines for LPC analysis can be found at
wolffd@0 50 % http://www.ee.ic.ac.uk/hp/staff/dmb/voicebox/voicebox.html
wolffd@0 51
wolffd@0 52
wolffd@0 53
wolffd@0 54 if (nargin < 5), fr = 20; end;
wolffd@0 55 if (nargin < 6), fs = 30; end;
wolffd@0 56 if (nargin < 7), preemp = .9378; end;
wolffd@0 57
wolffd@0 58 msfs = round(sr*fs/1000);
wolffd@0 59 msfr = round(sr*fr/1000);
wolffd@0 60 msoverlap = msfs - msfr;
wolffd@0 61 ramp = [0:1/(msoverlap-1):1]';
wolffd@0 62 [L1 nframe] = size(aCoeff); % L1 = 1+number of LPC coeffs
wolffd@0 63
wolffd@0 64 [row col] = size(source);
wolffd@0 65 if(row==1 | col==1) % continous stream; must be
wolffd@0 66 % windowed
wolffd@0 67 postFilter = 0; duration = length(source); frameIndex = 1;
wolffd@0 68 for sampleIndex=1:msfr:duration-msfs+1
wolffd@0 69 resid(:,frameIndex) = source(sampleIndex:(sampleIndex+msfs-1))';
wolffd@0 70 frameIndex = frameIndex+1;
wolffd@0 71 end
wolffd@0 72 else
wolffd@0 73 postFilter = 1; resid = source;
wolffd@0 74 end
wolffd@0 75
wolffd@0 76 [row col] = size(resid);
wolffd@0 77 %if ~(col==nframe)
wolffd@0 78 % error('synLPC: numbers of LPC frames and source frames do not match');
wolffd@0 79 if col<nframe
wolffd@0 80 nframe=col;
wolffd@0 81 end
wolffd@0 82
wolffd@0 83 for frameIndex=1:nframe
wolffd@0 84 % Calculate the filter response
wolffd@0 85 % by evaluating the z-transform
wolffd@0 86 % if 1
wolffd@0 87 % gain=0;
wolffd@0 88 % cft=0:(1/255):1;
wolffd@0 89 % for index=1:L1-1
wolffd@0 90 % gain = gain + aCoeff(index,frameIndex)*exp(-i*2*pi*cft).^index;
wolffd@0 91 % end
wolffd@0 92 % gain = abs(1./gain);
wolffd@0 93 % spec(:,frameIndex) = 20*log10(gain(1:128))';
wolffd@0 94 % plot(20*log10(gain));
wolffd@0 95 % title(frameIndex);
wolffd@0 96 % drawnow;
wolffd@0 97 % end
wolffd@0 98
wolffd@0 99 % Calculate the filter response
wolffd@0 100 % from the filter's impulse
wolffd@0 101 % response (to check above).
wolffd@0 102 % if 0
wolffd@0 103 % impulseResponse = filter(1, aCoeff(:,frameIndex), [1 zeros(1,255)]);
wolffd@0 104 % freqResp = 20*log10(abs(fft(impulseResponse)));
wolffd@0 105 % plot(freqResp);
wolffd@0 106 % end
wolffd@0 107
wolffd@0 108
wolffd@0 109 A = aCoeff(:,frameIndex);
wolffd@0 110 residFrame = resid(:,frameIndex)*G(frameIndex);
wolffd@0 111 synFrame = filter(1, A', residFrame); % synthesize speech from LPC
wolffd@0 112 % coeffs
wolffd@0 113 if(frameIndex==1) % add synthesize frames using a
wolffd@0 114 synWave = synFrame(1:msfr); % trapezoidal window
wolffd@0 115 else
wolffd@0 116 synWave = [synWave; overlap+synFrame(1:msoverlap).*ramp; ...
wolffd@0 117 synFrame(msoverlap+1:msfr)];
wolffd@0 118 end
wolffd@0 119 if(frameIndex==nframe)
wolffd@0 120 synWave = [synWave; synFrame(msfr+1:msfs)];
wolffd@0 121 else
wolffd@0 122 overlap = synFrame(msfr+1:msfs).*flipud(ramp);
wolffd@0 123 end
wolffd@0 124 %length(synWave)
wolffd@0 125 end;
wolffd@0 126
wolffd@0 127 if(postFilter)
wolffd@0 128 synWave = filter(1, [1 -preemp], synWave);
wolffd@0 129 end
wolffd@0 130