Mercurial > hg > camir-aes2014

annotate core/magnatagatune/MTTMixedFeatureStober11.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 classdef MTTMixedFeatureStober11 < MTTAudioFeature & handle
wolffd@0 2 % ---
wolffd@0 3 % This Class contains
wolffd@0 4 % features are extracted
wolffd@0 5 % as described in Slaney 08 - LEARNING A METRIC FOR MUSIC SIMILARITY
wolffd@0 6 %
wolffd@0 7 % The usual workflow for these features constist of three steps
wolffd@0 8 % 1. extract: extracts the basic single-file dependent features
wolffd@0 9 % 2. define_global_transform: calculates the global feature
wolffd@0 10 % transformation parameters
wolffd@0 11 % 3. finalise: applies the common transformations to a specific feature
wolffd@0 12 % ---
wolffd@0 13
wolffd@0 14 properties(Constant = true)
wolffd@0 15
wolffd@0 16 % svn hook
wolffd@0 17 my_revision = str2double(substr('$Rev$', 5, -1));
wolffd@0 18 end
wolffd@0 19
wolffd@0 20 properties
wolffd@0 21 % ---
wolffd@0 22 % Set default parameters
wolffd@0 23 % ---
wolffd@0 24 my_params = struct(...
wolffd@0 25 'stob_lowaudio', 1, ...
wolffd@0 26 'stob_highaudio', 1, ... %
wolffd@0 27 'stob_tags', 1, ... %
wolffd@0 28 'stob_norm', 1 ...
wolffd@0 29 );
wolffd@0 30 end
wolffd@0 31
wolffd@0 32 % ---
wolffd@0 33 % member functions
wolffd@0 34 % ---
wolffd@0 35 methods
wolffd@0 36
wolffd@0 37 % ---
wolffd@0 38 % constructor: pointer to feature in database
wolffd@0 39 % ---
wolffd@0 40 function feature = MTTMixedFeatureStober11(varargin)
wolffd@0 41
wolffd@0 42 feature = feature@MTTAudioFeature(varargin{:});
wolffd@0 43
wolffd@0 44 end
wolffd@0 45 % ---
wolffd@0 46 % extract feature data from raw audio features
wolffd@0 47 % ---
wolffd@0 48 function data = extract(feature, clip)
wolffd@0 49 % ---
wolffd@0 50 % get features. this includes possible
wolffd@0 51 % local normalisations
wolffd@0 52 % ---
wolffd@0 53
wolffd@0 54 global globalvars;
wolffd@0 55 global stobbase;
wolffd@0 56
wolffd@0 57 if isempty(stobbase);
wolffd@0 58 stobbase = load('features_stober');
wolffd@0 59 end
wolffd@0 60
wolffd@0 61 % ---
wolffd@0 62 % NOTE: we define feature sets which are included /
wolffd@0 63 % excluded according to the specified parameters
wolffd@0 64 % ---
wolffd@0 65
wolffd@0 66 lowAudio = {'pitchMean', 'pitchSdev', 'timbreMean', 'timbreSdev'};
wolffd@0 67
wolffd@0 68 % highAudio features more or less correspond to slaney 08 features
wolffd@0 69 hiAudio = {'energy','key','loudness','timeSignature',...
wolffd@0 70 'danceability', 'mode', 'tempo'};
wolffd@0 71
wolffd@0 72 metadat = {'tags'};
wolffd@0 73
wolffd@0 74 allowedFeat = {};
wolffd@0 75
wolffd@0 76 % Select the features to keep
wolffd@0 77 if feature(1).my_params.stob_lowaudio
wolffd@0 78 allowedFeat = {allowedFeat{:}, lowAudio{:}};
wolffd@0 79 end
wolffd@0 80 if feature(1).my_params.stob_highaudio
wolffd@0 81 allowedFeat = {allowedFeat{:}, hiAudio{:}};
wolffd@0 82 end
wolffd@0 83 if feature(1).my_params.stob_tags
wolffd@0 84 allowedFeat = {allowedFeat{:}, metadat{:}};
wolffd@0 85 end
wolffd@0 86
wolffd@0 87 % get the actual clip id
wolffd@0 88 idx = find(stobbase.clipIds == clip.id);
wolffd@0 89
wolffd@0 90 % ---
wolffd@0 91 % NOTE: we just copy everything in a big matrix and then
wolffd@0 92 % normalise the data later
wolffd@0 93 % ---
wolffd@0 94 data.vector_info = {};
wolffd@0 95 data.stobraw = [];
wolffd@0 96 fields = fieldnames(stobbase);
wolffd@0 97 for i = 1:numel(fields)
wolffd@0 98
wolffd@0 99 % skip clip ID field
wolffd@0 100 if strcmp(fields{i},'clipIds');
wolffd@0 101 continue;
wolffd@0 102 end
wolffd@0 103
wolffd@0 104 % skip unwanted features
wolffd@0 105 if isempty(strcellfind(allowedFeat, fields{i}))
wolffd@0 106 continue;
wolffd@0 107 end
wolffd@0 108
wolffd@0 109 % ---
wolffd@0 110 % TODO: special case for tag features, including
wolffd@0 111 % the tag names
wolffd@0 112 % ---
wolffd@0 113
wolffd@0 114 % put field info into right position
wolffd@0 115 data.vector_info{numel(data.stobraw)+1} = fields{i};
wolffd@0 116
wolffd@0 117 % add data to feature
wolffd@0 118 if size(stobbase.(fields{i}),1) == 1
wolffd@0 119 data.stobraw(end+1) = stobbase.(fields{i})(idx);
wolffd@0 120 else
wolffd@0 121 % concatenate vector
wolffd@0 122 tmpdat = stobbase.(fields{i})(idx,:);
wolffd@0 123 data.stobraw(end+1:end+numel(tmpdat)) = tmpdat;
wolffd@0 124 end
wolffd@0 125 end
wolffd@0 126 % padd further info struct
wolffd@0 127 data.vector_info(end+1:numel(data.stobraw)) =...
wolffd@0 128 cell(numel(data.stobraw) - numel(data.vector_info) , 1);
wolffd@0 129
wolffd@0 130 % ---
wolffd@0 131 % prepare field for final features
wolffd@0 132 % ---
wolffd@0 133 data.final.vector = [];
wolffd@0 134 data.final.vector_info = struct();
wolffd@0 135 data.final.dim = 0;
wolffd@0 136
wolffd@0 137 % save info data
wolffd@0 138 data.info.type = 'MTTMixedFeatureStober11';
wolffd@0 139 data.info.owner_id = clip.id;
wolffd@0 140 data.info.creatorrev = feature.my_revision;
wolffd@0 141
wolffd@0 142 % save parameters
wolffd@0 143 data.info.params = feature.my_params;
wolffd@0 144 end
wolffd@0 145
wolffd@0 146 function define_global_transform(features)
wolffd@0 147 % calculate and set normalization factors from the group of
wolffd@0 148 % input features. These features will be set for the full database
wolffd@0 149
wolffd@0 150 final = zeros(numel(features(1).data.stobraw), numel(features));
wolffd@0 151 for i = 1:numel(features)
wolffd@0 152 if ~isempty(features(i).data.stobraw)
wolffd@0 153 final(:,i) = features(i).data.stobraw';
wolffd@0 154 end
wolffd@0 155 end
wolffd@0 156
wolffd@0 157 if features(1).my_params.stob_norm
wolffd@0 158 if numel(features) == 1
wolffd@0 159 error ('Insert feature array for this method, or set normalisation to 0');
wolffd@0 160 end
wolffd@0 161
wolffd@0 162 % ---
wolffd@0 163 % here, we only need to define the post-normalisation
wolffd@0 164 % ---
wolffd@0 165 [final, pstd] = mapminmax(final,0,1);
wolffd@0 166 common.stobstats.post_norm = pstd;
wolffd@0 167
wolffd@0 168 % ---
wolffd@0 169 % NOTE: whitening as in slaney??
wolffd@0 170 % Would make reading the
wolffd@0 171 % mahal matrices really hard
wolffd@0 172 % ---
wolffd@0 173
wolffd@0 174 features(1).my_db.set_common(common);
wolffd@0 175
wolffd@0 176 else
wolffd@0 177
wolffd@0 178 features(1).my_db.set_common([1]);
wolffd@0 179 end
wolffd@0 180
wolffd@0 181 % save the normalised features straight away!
wolffd@0 182 features.finalise(final);
wolffd@0 183 end
wolffd@0 184
wolffd@0 185
wolffd@0 186 function finalise(features, final)
wolffd@0 187 % applies a final transformation and
wolffd@0 188 % collects the information of this feature within a single vector
wolffd@0 189 % see info for types in specific dimensions
wolffd@0 190 % check if features have been finalised already
wolffd@0 191
wolffd@0 192 % ---
wolffd@0 193 % set feature labelling
wolffd@0 194 % ---
wolffd@0 195
wolffd@0 196 info = {};
wolffd@0 197
wolffd@0 198 % ---
wolffd@0 199 % construct resulting feature vector out of features
wolffd@0 200 % ---
wolffd@0 201 if nargin == 2 && isempty(final)
wolffd@0 202
wolffd@0 203 % the final vector etc already are set to zero;
wolffd@0 204 return;
wolffd@0 205
wolffd@0 206 elseif nargin == 2 && (numel(features) == size(final, 2))
wolffd@0 207 % the features have already been preassembled
wolffd@0 208
wolffd@0 209 for i = 1:numel(features)
wolffd@0 210
wolffd@0 211 % check for neccesary parameters
wolffd@0 212 if isempty(features(i).my_db.commondb)
wolffd@0 213
wolffd@0 214 error('Define the global transformation first')
wolffd@0 215 return;
wolffd@0 216 end
wolffd@0 217
wolffd@0 218 features(i).data.final.vector = final(:,i);
wolffd@0 219 features(i).data.final.dim = size(final,1);
wolffd@0 220
wolffd@0 221 % fill up info struct and append to feature
wolffd@0 222 features(i).data.final.vector_info.labels = ...
wolffd@0 223 features(i).data.vector_info;
wolffd@0 224 end
wolffd@0 225 else
wolffd@0 226 % ---
wolffd@0 227 % if features have been added after gettin gnormalisation
wolffd@0 228 % parameters, ther should be still an option to include
wolffd@0 229 % them
wolffd@0 230 % ---
wolffd@0 231
wolffd@0 232 for i = 1:numel(features)
wolffd@0 233
wolffd@0 234 % check for neccesary parameters
wolffd@0 235 if isempty(features(i).my_db.commondb)
wolffd@0 236
wolffd@0 237 error('Define the global transformation first')
wolffd@0 238 return;
wolffd@0 239 end
wolffd@0 240
wolffd@0 241 final = features(i).data.stobraw';
wolffd@0 242
wolffd@0 243 if features(1).my_params.stob_norm == 1
wolffd@0 244
wolffd@0 245 [final] = mapminmax('apply', final, features(1).common.stobstats.post_norm);
wolffd@0 246 end
wolffd@0 247
wolffd@0 248 features(i).data.final.vector = final;
wolffd@0 249 features(i).data.final.dim = size(final,1);
wolffd@0 250
wolffd@0 251 % fill up info struct and append to feature
wolffd@0 252 features(i).data.final.vector_info.labels = ...
wolffd@0 253 features(i).data.vector_info;
wolffd@0 254 end
wolffd@0 255
wolffd@0 256 end
wolffd@0 257
wolffd@0 258 % ---
wolffd@0 259 % TODO: Maybe delete more basic features again at this point?
wolffd@0 260 % ---
wolffd@0 261 end
wolffd@0 262
wolffd@0 263 % ---
wolffd@0 264 % destructor: do we really want to remove this
wolffd@0 265 % from the database? No, but
wolffd@0 266 % TODO: create marker for unused objects in db, and a cleanup
wolffd@0 267 % function
wolffd@0 268 % ---
wolffd@0 269 function delete(feature)
wolffd@0 270
wolffd@0 271 end
wolffd@0 272 end
wolffd@0 273 end