Mercurial > hg > camir-aes2014

annotate toolboxes/MIRtoolbox1.3.2/MIRToolbox/@mirpitch/mirpitch.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 varargout = mirpitch(orig,varargin)
wolffd@0 2 % p = mirpitch(x) evaluates the pitch frequencies (in Hz).
wolffd@0 3 % Specification of the method(s) for pitch estimation (these methods can
wolffd@0 4 % be combined):
wolffd@0 5 % mirpitch(...,'Autocor') computes an autocorrelation function
wolffd@0 6 % (Default method)
wolffd@0 7 % mirpitch(...'Enhanced',a) computes enhanced autocorrelation
wolffd@0 8 % (see help mirautocor)
wolffd@0 9 % toggled on by default
wolffd@0 10 % mirpitch(...,'Compress',k) performs magnitude compression
wolffd@0 11 % (see help mirautocor)
wolffd@0 12 % mirpitch(...,fb) specifies a type of filterbank.
wolffd@0 13 % Possible values:
wolffd@0 14 % fb = 'NoFilterBank': no filterbank decomposition
wolffd@0 15 % fb = '2Channels' (default value)
wolffd@0 16 % fb = 'Gammatone'
wolffd@0 17 % mirpitch(...,'AutocorSpectrum') computes the autocorrelation of
wolffd@0 18 % the FFT spectrum
wolffd@0 19 % mirpitch(...,'Cepstrum') computes the cepstrum
wolffd@0 20 % Alternatively, an autocorrelation or a cepstrum can be directly
wolffd@0 21 % given as first argument of the mirpitch function.
wolffd@0 22 % Peak picking options:
wolffd@0 23 % mirpitch(...,'Total',m) selects the m best pitches.
wolffd@0 24 % Default value: m = Inf, no limit is set concerning the number
wolffd@0 25 % of pitches to be detected.
wolffd@0 26 % mirpitch(...,'Mono') corresponds to morpitch(...,'Total',1)
wolffd@0 27 % mirpitch(...,'Min',mi) indicates the lowest frequency taken into
wolffd@0 28 % consideration.
wolffd@0 29 % Default value: 75 Hz. (Praat)
wolffd@0 30 % mirpitch(...,'Max',ma) indicates the highest frequency taken into
wolffd@0 31 % consideration.
wolffd@0 32 % Default value: 2400 Hz. Because there seems to be some problems
wolffd@0 33 % with higher frequency, due probably to the absence of
wolffd@0 34 % pre-whitening in our implementation of Tolonen and Karjalainen
wolffd@0 35 % approach (used by default, cf. below).
wolffd@0 36 % mirpitch(...,'Contrast',thr) specifies a threshold value.
wolffd@0 37 % (see help peaks)
wolffd@0 38 % Default value: thr = .1
wolffd@0 39 % mirpitch(...,'Order',o) specifies the ordering for the peak picking.
wolffd@0 40 % Default value: o = 'Amplitude'.
wolffd@0 41 % Alternatively, the result of a mirpeaks computation can be directly
wolffd@0 42 % given as first argument of the mirpitch function.
wolffd@0 43 % Post-processing options:
wolffd@0 44 % mirpitch(...,'Sum','no') does not sum back the channels at the end
wolffd@0 45 % of the computation. The resulting pitch information remains
wolffd@0 46 % therefore decomposed into several channels.
wolffd@0 47 % mirpitch(...,'Median') performs a median filtering of the pitch
wolffd@0 48 % curve. When several pitches are extracted in each frame, the
wolffd@0 49 % pitch curve contains the best peak of each successive frame.
wolffd@0 50 % mirpitch(...,'Stable',th,n) remove pitch values when the difference
wolffd@0 51 % (or more precisely absolute logarithmic quotient) with the
wolffd@0 52 % n precedent frames exceeds the threshold th.
wolffd@0 53 % if th is not specified, the default value .1 is used
wolffd@0 54 % if n is not specified, the default value 3 is used
wolffd@0 55 % mirpitch(...'Reso',r) removes peaks whose distance to one or
wolffd@0 56 % several higher peaks is lower than a given threshold.
wolffd@0 57 % Possible value for the threshold r:
wolffd@0 58 % 'SemiTone': ratio between the two peak positions equal to
wolffd@0 59 % 2^(1/12)
wolffd@0 60 % mirpitch(...,'Frame',l,h) orders a frame decomposition of window
wolffd@0 61 % length l (in seconds) and hop factor h, expressed relatively to
wolffd@0 62 % the window length. For instance h = 1 indicates no overlap.
wolffd@0 63 % Default values: l = 46.4 ms and h = 10 ms (Tolonen and
wolffd@0 64 % Karjalainen, 2000)
wolffd@0 65 % Preset model:
wolffd@0 66 % mirpitch(...,'Tolonen') implements (part of) the model proposed in
wolffd@0 67 % (Tolonen & Karjalainen, 2000). It is equivalent to
wolffd@0 68 % mirpitch(...,'Enhanced',2:10,'Generalized',.67,'2Channels')
wolffd@0 69 % [p,a] = mirpitch(...) also displays the result of the method chosen for
wolffd@0 70 % pitch estimation, and shows in particular the peaks corresponding
wolffd@0 71 % to the pitch values.
wolffd@0 72 % p = mirpitch(f,a,<r>) creates a mirpitch object based on the frequencies
wolffd@0 73 % specified in f and the related amplitudes specified in a, using a
wolffd@0 74 % frame sampling rate of r Hz (set by default to 100 Hz).
wolffd@0 75 %
wolffd@0 76 % T. Tolonen, M. Karjalainen, "A Computationally Efficient Multipitch
wolffd@0 77 % Analysis Model", IEEE TRANSACTIONS ON SPEECH AND AUDIO PROCESSING,
wolffd@0 78 % VOL. 8, NO. 6, NOVEMBER 2000
wolffd@0 79
wolffd@0 80 ac.key = 'Autocor';
wolffd@0 81 ac.type = 'Boolean';
wolffd@0 82 ac.default = 0;
wolffd@0 83 option.ac = ac;
wolffd@0 84
wolffd@0 85 enh.key = 'Enhanced';
wolffd@0 86 enh.type = 'Integer';
wolffd@0 87 enh.default = 2:10;
wolffd@0 88 option.enh = enh;
wolffd@0 89
wolffd@0 90 filtertype.type = 'String';
wolffd@0 91 filtertype.choice = {'NoFilterBank','2Channels','Gammatone'};
wolffd@0 92 filtertype.default = '2Channels';
wolffd@0 93 option.filtertype = filtertype;
wolffd@0 94
wolffd@0 95 gener.key = {'Generalized','Compress'};
wolffd@0 96 gener.type = 'Integer';
wolffd@0 97 gener.default = .5;
wolffd@0 98 option.gener = gener;
wolffd@0 99
wolffd@0 100 as.key = 'AutocorSpectrum';
wolffd@0 101 as.type = 'Boolean';
wolffd@0 102 as.default = 0;
wolffd@0 103 option.as = as;
wolffd@0 104
wolffd@0 105 s.key = 'Spectrum';
wolffd@0 106 s.type = 'Boolean';
wolffd@0 107 s.default = 0;
wolffd@0 108 option.s = s;
wolffd@0 109
wolffd@0 110 ce.key = 'Cepstrum';
wolffd@0 111 ce.type = 'Boolean';
wolffd@0 112 ce.default = 0;
wolffd@0 113 option.ce = ce;
wolffd@0 114
wolffd@0 115 %% peak picking options
wolffd@0 116
wolffd@0 117 m.key = 'Total';
wolffd@0 118 m.type = 'Integer';
wolffd@0 119 m.default = Inf;
wolffd@0 120 option.m = m;
wolffd@0 121
wolffd@0 122 multi.key = 'Multi';
wolffd@0 123 multi.type = 'Boolean';
wolffd@0 124 multi.default = 0;
wolffd@0 125 option.multi = multi;
wolffd@0 126
wolffd@0 127 mono.key = 'Mono';
wolffd@0 128 mono.type = 'Boolean';
wolffd@0 129 mono.default = 0;
wolffd@0 130 option.mono = mono;
wolffd@0 131
wolffd@0 132 mi.key = 'Min';
wolffd@0 133 mi.type = 'Integer';
wolffd@0 134 mi.default = 75;
wolffd@0 135 option.mi = mi;
wolffd@0 136
wolffd@0 137 ma.key = 'Max';
wolffd@0 138 ma.type = 'Integer';
wolffd@0 139 ma.default = 2400;
wolffd@0 140 option.ma = ma;
wolffd@0 141
wolffd@0 142 thr.key = 'Contrast';
wolffd@0 143 thr.type = 'Integer';
wolffd@0 144 thr.default = .1;
wolffd@0 145 option.thr = thr;
wolffd@0 146
wolffd@0 147 order.key = 'Order';
wolffd@0 148 order.type = 'String';
wolffd@0 149 order.choice = {'Amplitude','Abscissa'};
wolffd@0 150 order.default = 'Amplitude';
wolffd@0 151 option.order = order;
wolffd@0 152
wolffd@0 153 reso.key = 'Reso';
wolffd@0 154 reso.type = 'String';
wolffd@0 155 reso.choice = {0,'SemiTone'};
wolffd@0 156 reso.default = 0;
wolffd@0 157 option.reso = reso;
wolffd@0 158
wolffd@0 159 track.key = 'Track'; % Not used yet
wolffd@0 160 track.type = 'Boolean';
wolffd@0 161 track.default = 0;
wolffd@0 162 option.track = track;
wolffd@0 163
wolffd@0 164 %% post-processing options
wolffd@0 165
wolffd@0 166 stable.key = 'Stable';
wolffd@0 167 stable.type = 'Integer';
wolffd@0 168 stable.number = 2;
wolffd@0 169 stable.default = [Inf 0];
wolffd@0 170 stable.keydefault = [.1 3];
wolffd@0 171 option.stable = stable;
wolffd@0 172
wolffd@0 173 median.key = 'Median';
wolffd@0 174 median.type = 'Integer';
wolffd@0 175 median.default = 0;
wolffd@0 176 median.keydefault = .1;
wolffd@0 177 option.median = median;
wolffd@0 178
wolffd@0 179 frame.key = 'Frame';
wolffd@0 180 frame.type = 'Integer';
wolffd@0 181 frame.number = 2;
wolffd@0 182 frame.default = [0 0];
wolffd@0 183 frame.keydefault = [NaN NaN];
wolffd@0 184 option.frame = frame;
wolffd@0 185
wolffd@0 186 sum.key = 'Sum';
wolffd@0 187 sum.type = 'Boolean';
wolffd@0 188 sum.default = 1;
wolffd@0 189 option.sum = sum;
wolffd@0 190
wolffd@0 191 %% preset model
wolffd@0 192
wolffd@0 193 tolo.key = 'Tolonen';
wolffd@0 194 tolo.type = 'Boolean';
wolffd@0 195 tolo.default = 0;
wolffd@0 196 option.tolo = tolo;
wolffd@0 197
wolffd@0 198 specif.option = option;
wolffd@0 199 specif.chunkframebefore = 1;
wolffd@0 200
wolffd@0 201 if isnumeric(orig)
wolffd@0 202 if nargin<3
wolffd@0 203 f = 100;
wolffd@0 204 else
wolffd@0 205 f = varargin{2};
wolffd@0 206 end
wolffd@0 207 fp = (0:size(orig,1)-1)/f;
wolffd@0 208 fp = [fp;fp+1/f];
wolffd@0 209 p.amplitude = {{varargin{1}'}};
wolffd@0 210 s = mirscalar([],'Data',{{orig'}},'Title','Pitch','Unit','Hz',...
wolffd@0 211 'FramePos',{{fp}},'Sampling',f,'Name',{inputname(1)});
wolffd@0 212 p = class(p,'mirpitch',s);
wolffd@0 213 varargout = {p};
wolffd@0 214 else
wolffd@0 215 varargout = mirfunction(@mirpitch,orig,varargin,nargout,specif,@init,@main);
wolffd@0 216 end
wolffd@0 217
wolffd@0 218
wolffd@0 219
wolffd@0 220 function [y type] = init(orig,option)
wolffd@0 221 if option.tolo
wolffd@0 222 option.enh = 2:10;
wolffd@0 223 option.gener = .67;
wolffd@0 224 option.filtertype = '2Channels';
wolffd@0 225 end
wolffd@0 226 if not(option.ac) && not(option.as) && not(option.ce) && not(option.s)
wolffd@0 227 option.ac = 1;
wolffd@0 228 end
wolffd@0 229 if isnan(option.frame.length.val)
wolffd@0 230 option.frame.length.val = .0464;
wolffd@0 231 end
wolffd@0 232 if isnan(option.frame.hop.val)
wolffd@0 233 option.frame.hop.val = .01;
wolffd@0 234 option.frame.hop.unit = 's';
wolffd@0 235 end
wolffd@0 236 if isamir(orig,'mirscalar') || haspeaks(orig)
wolffd@0 237 y = orig;
wolffd@0 238 else
wolffd@0 239 if isamir(orig,'mirautocor')
wolffd@0 240 y = mirautocor(orig,'Min',option.mi,'Hz','Max',option.ma,'Hz','Freq');
wolffd@0 241 elseif isamir(orig,'mircepstrum')
wolffd@0 242 y = orig;
wolffd@0 243 elseif isamir(orig,'mirspectrum')
wolffd@0 244 if not(option.as) && not(option.ce) && not(option.s)
wolffd@0 245 option.ce = 1;
wolffd@0 246 end
wolffd@0 247 if option.as
wolffd@0 248 y = mirautocor(orig,...
wolffd@0 249 'Min',option.mi,'Hz','Max',option.ma,'Hz');
wolffd@0 250 end
wolffd@0 251 if option.ce
wolffd@0 252 ce = mircepstrum(orig,'freq',...
wolffd@0 253 'Min',option.mi,'Hz','Max',option.ma,'Hz');
wolffd@0 254 if option.as
wolffd@0 255 y = y*ce;
wolffd@0 256 else
wolffd@0 257 y = ce;
wolffd@0 258 end
wolffd@0 259 end
wolffd@0 260 else
wolffd@0 261 if option.ac
wolffd@0 262 x = orig;
wolffd@0 263 if not(strcmpi(option.filtertype,'NoFilterBank'))
wolffd@0 264 x = mirfilterbank(x,option.filtertype);
wolffd@0 265 end
wolffd@0 266 x = mirframenow(x,option);
wolffd@0 267 y = mirautocor(x,'Generalized',option.gener,...
wolffd@0 268 'Min',option.mi,'Hz','Max',option.ma,'Hz');
wolffd@0 269 if option.sum
wolffd@0 270 y = mirsummary(y);
wolffd@0 271 end
wolffd@0 272 y = mirautocor(y,'Enhanced',option.enh,'Freq');
wolffd@0 273 end
wolffd@0 274 if option.as || option.ce || option.s
wolffd@0 275 x = mirframenow(orig,option);
wolffd@0 276 y = mirspectrum(x);
wolffd@0 277 if option.as
wolffd@0 278 as = mirautocor(y,...
wolffd@0 279 'Min',option.mi,'Hz','Max',option.ma,'Hz');
wolffd@0 280 if option.ac
wolffd@0 281 y = y*as;
wolffd@0 282 else
wolffd@0 283 y = as;
wolffd@0 284 end
wolffd@0 285 end
wolffd@0 286 if option.ce
wolffd@0 287 ce = mircepstrum(y,'freq',...
wolffd@0 288 'Min',option.mi,'Hz','Max',option.ma,'Hz');
wolffd@0 289 if option.ac || option.as
wolffd@0 290 y = y*ce;
wolffd@0 291 else
wolffd@0 292 y = ce;
wolffd@0 293 end
wolffd@0 294 end
wolffd@0 295 end
wolffd@0 296 end
wolffd@0 297 end
wolffd@0 298 type = {'mirpitch',mirtype(y)};
wolffd@0 299
wolffd@0 300
wolffd@0 301 function o = main(x,option,postoption)
wolffd@0 302 if option.multi && option.m == 1
wolffd@0 303 option.m = Inf;
wolffd@0 304 end
wolffd@0 305 if option.mono && option.m == Inf
wolffd@0 306 option.m = 1;
wolffd@0 307 end
wolffd@0 308 if iscell(x)
wolffd@0 309 x = x{1};
wolffd@0 310 end
wolffd@0 311 if not(isa(x,'mirpitch'))
wolffd@0 312 x = mirpeaks(x,'Total',option.m,'Track',option.track,...
wolffd@0 313 'Contrast',option.thr,'Threshold',.4,...
wolffd@0 314 'Reso',option.reso,'NoBegin','NoEnd',...
wolffd@0 315 'Order',option.order);
wolffd@0 316 end
wolffd@0 317 if isa(x,'mirscalar')
wolffd@0 318 pf = get(x,'Data');
wolffd@0 319 else
wolffd@0 320 pf = get(x,'PeakPrecisePos');
wolffd@0 321 pa = get(x,'PeakPreciseVal');
wolffd@0 322 end
wolffd@0 323 fp = get(x,'FramePos');
wolffd@0 324 if option.stable(1) < Inf
wolffd@0 325 for i = 1:length(pf)
wolffd@0 326 for j = 1:length(pf{i})
wolffd@0 327 for k = 1:size(pf{i}{j},3)
wolffd@0 328 for l = size(pf{i}{j},2):-1:option.stable(2)+1
wolffd@0 329 for m = length(pf{i}{j}{1,l,k}):-1:1
wolffd@0 330 found = 0;
wolffd@0 331 for h = 1:option.stable(2)
wolffd@0 332 for n = 1:length(pf{i}{j}{1,l-h,k})
wolffd@0 333 if abs(log10(pf{i}{j}{1,l,k}(m) ...
wolffd@0 334 /pf{i}{j}{1,l-h,k}(n))) ...
wolffd@0 335 < option.stable(1)
wolffd@0 336 found = 1;
wolffd@0 337 end
wolffd@0 338 end
wolffd@0 339 end
wolffd@0 340 if not(found)
wolffd@0 341 pf{i}{j}{1,l,k}(m) = [];
wolffd@0 342 end
wolffd@0 343 end
wolffd@0 344 pf{i}{j}{1,1,k} = zeros(1,0);
wolffd@0 345 end
wolffd@0 346 end
wolffd@0 347 end
wolffd@0 348 end
wolffd@0 349 end
wolffd@0 350 if option.median
wolffd@0 351 sr = get(x,'Sampling');
wolffd@0 352 for i = 1:length(pf)
wolffd@0 353 for j = 1:length(pf{i})
wolffd@0 354 if size(fp{i}{j},2) > 1
wolffd@0 355 npf = zeros(size(pf{i}{j}));
wolffd@0 356 for k = 1:size(pf{i}{j},3)
wolffd@0 357 for l = 1:size(pf{i}{j},2)
wolffd@0 358 if isempty(pf{i}{j}{1,l,k})
wolffd@0 359 npf(1,l,k) = NaN;
wolffd@0 360 else
wolffd@0 361 npf(1,l,k) = pf{i}{j}{1,l,k}(1);
wolffd@0 362 end
wolffd@0 363 end
wolffd@0 364 end
wolffd@0 365 pf{i}{j} = medfilt1(npf,...
wolffd@0 366 round(option.median/(fp{i}{j}(1,2)-fp{i}{j}(1,1))));
wolffd@0 367 end
wolffd@0 368 end
wolffd@0 369 end
wolffd@0 370 end
wolffd@0 371 if isa(x,'mirscalar')
wolffd@0 372 p.amplitude = 0;
wolffd@0 373 else
wolffd@0 374 p.amplitude = pa;
wolffd@0 375 end
wolffd@0 376 s = mirscalar(x,'Data',pf,'Title','Pitch','Unit','Hz');
wolffd@0 377 p = class(p,'mirpitch',s);
wolffd@0 378 o = {p,x};