Mercurial > hg > camir-aes2014

annotate toolboxes/MIRtoolbox1.3.2/AuditoryToolbox/MakeVowel.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 y=MakeVowel(len, pitch, sampleRate, f1, f2, f3)
wolffd@0 2 % MakeVowel(len, pitch [, sampleRate, f1, f2, f3]) - Make a vowel with
wolffd@0 3 % "len" samples and the given pitch. The sample rate defaults to
wolffd@0 4 % be 22254.545454 Hz (the native Mactinosh Sampling Rate). The
wolffd@0 5 % formant frequencies are f1, f2 & f3. Some common vowels are
wolffd@0 6 % Vowel f1 f2 f3
wolffd@0 7 % /a/ 730 1090 2440
wolffd@0 8 % /i/ 270 2290 3010
wolffd@0 9 % /u/ 300 870 2240
wolffd@0 10 %
wolffd@0 11 % The pitch variable can either be a scalar indicating the actual
wolffd@0 12 % pitch frequency, or an array of impulse locations. Using an
wolffd@0 13 % array of impulses allows this routine to compute vowels with
wolffd@0 14 % varying pitch.
wolffd@0 15 %
wolffd@0 16 % Alternatively, f1 can be replaced with one of the following strings
wolffd@0 17 % 'a', 'i', 'u' and the appropriate formant frequencies are
wolffd@0 18 % automatically selected.
wolffd@0 19 % Modified by R. Duda, 3/13/94
wolffd@0 20
wolffd@0 21 % (c) 1998 Interval Research Corporation
wolffd@0 22
wolffd@0 23 if nargin < 2,
wolffd@0 24 fprintf('Format: y = MakeVowel(len, pitch [, sampleRate, f1, f2, f3])\n');
wolffd@0 25 return;
wolffd@0 26 end;
wolffd@0 27
wolffd@0 28 if nargin < 6; f3 = 0; end;
wolffd@0 29 if nargin < 5; f2 = 0; end;
wolffd@0 30 if nargin < 4,
wolffd@0 31 f1 = 0;
wolffd@0 32 else
wolffd@0 33 if isstr(f1)
wolffd@0 34 if f1 == 'a' | f1 == '/a/'
wolffd@0 35 f1=730; f2=1090; f3=2440;
wolffd@0 36 elseif f1 == 'i' | f1 == '/i/'
wolffd@0 37 f1=270; f2=2290; f3=3010;
wolffd@0 38 elseif f1 == 'u' | f1 == '/u/'
wolffd@0 39 f1=300; f2=870; f3=2240;
wolffd@0 40 end
wolffd@0 41 end;
wolffd@0 42 end;
wolffd@0 43
wolffd@0 44 if nargin < 3,
wolffd@0 45 sampleRate = 22254.545454;
wolffd@0 46 elseif sampleRate < 1000, % Apparently for test purposes
wolffd@0 47 sampleRate = 22254.545454;
wolffd@0 48 end;
wolffd@0 49
wolffd@0 50 % GlottalPulses(pitch, fs, len) - Generate a stream of
wolffd@0 51 % glottal pulses with the given pitch (in Hz) and sampling
wolffd@0 52 % frequency (sampleRate). A vector of the requested length is returned.
wolffd@0 53 y=zeros(1,len);
wolffd@0 54 if length(pitch) > 1, % If true, use to determine points
wolffd@0 55 points=pitch; % Check for valid sequence of points
wolffd@0 56 if any(points~=sort(points)),
wolffd@0 57 error('Values in pitch array must be in ascending order.')
wolffd@0 58 end;
wolffd@0 59 if points(1) < 1,
wolffd@0 60 error('Values in pitch array cannot be less than 1.');
wolffd@0 61 end;
wolffd@0 62 kmax=sum(points <= len);
wolffd@0 63 if kmax == 0,
wolffd@0 64 error('All values in pitch array exceed "len"; none should.');
wolffd@0 65 elseif kmax < length(points),
wolffd@0 66 fprintf('Some values in pitch array exceed "len"; truncating.\n');
wolffd@0 67 points=points(1:kmax);
wolffd@0 68 end;
wolffd@0 69 else
wolffd@0 70 points=1:sampleRate/pitch:len;
wolffd@0 71 end;
wolffd@0 72 indices=floor(points);
wolffd@0 73
wolffd@0 74 % Use a triangular approximation to an impulse function. The important
wolffd@0 75 % part is to keep the total amplitude the same.
wolffd@0 76 y(indices) = (indices+1)-points;
wolffd@0 77 y(indices+1) = points-indices;
wolffd@0 78
wolffd@0 79 % GlottalFilter(x,fs) - Filter an impulse train and simulate the glottal
wolffd@0 80 % transfer function. The sampling interval (sampleRate) is given in Hz.
wolffd@0 81 % The filtering performed by this function is two first-order filters
wolffd@0 82 % at 250Hz.
wolffd@0 83 a = exp(-250*2*pi/sampleRate);
wolffd@0 84 %y=filter([1,0,-1],[1,-2*a,a*a],y); % Not as good as one below....
wolffd@0 85 y=filter([1],[1,0,-a*a],y);
wolffd@0 86
wolffd@0 87 % FormantFilter(input, f, fs) - Filter an input sequence to model one
wolffd@0 88 % formant in a speech signal. The formant frequency (in Hz) is given
wolffd@0 89 % by f and the bandwidth of the formant is a constant 50Hz. The
wolffd@0 90 % sampling frequency in Hz is given by fs.
wolffd@0 91 if f1 > 0
wolffd@0 92 cft = f1/sampleRate;
wolffd@0 93 bw = 50;
wolffd@0 94 q = f1/bw;
wolffd@0 95 rho = exp(-pi * cft / q);
wolffd@0 96 theta = 2 * pi * cft * sqrt(1-1/(4 * q*q));
wolffd@0 97 a2 = -2*rho*cos(theta);
wolffd@0 98 a3 = rho*rho;
wolffd@0 99 y=filter([1+a2+a3],[1,a2,a3],y);
wolffd@0 100 end;
wolffd@0 101
wolffd@0 102 % FormantFilter(input, f, fs) - Filter an input sequence to model one
wolffd@0 103 % formant in a speech signal. The formant frequency (in Hz) is given
wolffd@0 104 % by f and the bandwidth of the formant is a constant 50Hz. The
wolffd@0 105 % sampling frequency in Hz is given by fs.
wolffd@0 106 if f2 > 0
wolffd@0 107 cft = f2/sampleRate;
wolffd@0 108 bw = 50;
wolffd@0 109 q = f2/bw;
wolffd@0 110 rho = exp(-pi * cft / q);
wolffd@0 111 theta = 2 * pi * cft * sqrt(1-1/(4 * q*q));
wolffd@0 112 a2 = -2*rho*cos(theta);
wolffd@0 113 a3 = rho*rho;
wolffd@0 114 y=filter([1+a2+a3],[1,a2,a3],y);
wolffd@0 115 end;
wolffd@0 116
wolffd@0 117 % FormantFilter(input, f, fs) - Filter an input sequence to model one
wolffd@0 118 % formant in a speech signal. The formant frequency (in Hz) is given
wolffd@0 119 % by f and the bandwidth of the formant is a constant 50Hz. The
wolffd@0 120 % sampling frequency in Hz is given by fs.
wolffd@0 121 if f3 > 0
wolffd@0 122 cft = f3/sampleRate;
wolffd@0 123 bw = 50;
wolffd@0 124 q = f3/bw;
wolffd@0 125 rho = exp(-pi * cft / q);
wolffd@0 126 theta = 2 * pi * cft * sqrt(1-1/(4 * q*q));
wolffd@0 127 a2 = -2*rho*cos(theta);
wolffd@0 128 a3 = rho*rho;
wolffd@0 129 y=filter([1+a2+a3],[1,a2,a3],y);
wolffd@0 130 end;