wolffd@0: function features = xml_parse_mtt(file)
wolffd@0: % features = xml_parse_mtt(file)
wolffd@0: % 
wolffd@0: % parses the magnatagatune xml file to retrieve the audio
wolffd@0: % analysis results. The file isexpected to use the deprecated 
wolffd@0: % EchoNest XML format
wolffd@0: %
wolffd@0: % time always travels an the horizontal axis
wolffd@0: 
wolffd@0: % ---
wolffd@0: % TODO: Add Error checking to certan tags as chroma number, 
wolffd@0: %  loudness max etc.
wolffd@0: % ---
wolffd@0: 
wolffd@0: fdata = fileread(file);
wolffd@0: try
wolffd@0:     % ---
wolffd@0:     % TODO: check for other than the p-code exceptions
wolffd@0:     % ---
wolffd@0:     
wolffd@0:     data = xml_parseany(fdata);
wolffd@0: catch exception
wolffd@0: end
wolffd@0: %% ------------------------------------------------------------------------
wolffd@0: % ---
wolffd@0: % copy general track data
wolffd@0: % ---
wolffd@0: 
wolffd@0: fields = fieldnames(data.track{1}.ATTRIBUTE);
wolffd@0: for i = 1:length(fields)
wolffd@0:     features.(fields{i}) = str2double(data.track{1}.ATTRIBUTE.(fields{i}));
wolffd@0: end
wolffd@0: 
wolffd@0: %% ------------------------------------------------------------------------
wolffd@0: % ---
wolffd@0: % get harmonic analysis. this is stored in segments
wolffd@0: % ---
wolffd@0: dsegments = data.track{1}.segments{1}.segment;
wolffd@0: 
wolffd@0: for i = 1:numel(dsegments)
wolffd@0:     
wolffd@0:     % get start and duration
wolffd@0:     segments(i).start = str2double(dsegments{i}.ATTRIBUTE.start);
wolffd@0:     segments(i).duration = str2double(dsegments{i}.ATTRIBUTE.duration);
wolffd@0:     
wolffd@0:     % ---
wolffd@0:     % NOTE: for the chroma and mfcc features, we assume that the classes
wolffd@0:     %  are always saved and parsed in correct order, thus we can afford to 
wolffd@0:     %  refrain from saving the class number with the class
wolffd@0:     % ---
wolffd@0:     
wolffd@0:     % assemble chroma features
wolffd@0:     segments(i).pitches = zeros(12,1);
wolffd@0:     for j = 1:12
wolffd@0:         
wolffd@0:         segments(i).pitches(j) = str2double(...
wolffd@0:             dsegments{i}.pitches{1}.pitch{j}.CONTENT);
wolffd@0:     end
wolffd@0:     
wolffd@0:    % assemble mfcc features;
wolffd@0:     segments(i).timbre = zeros(numel(dsegments{i}.timbre{1}.coeff),1);
wolffd@0:     for j = 1:numel(dsegments{i}.timbre{1}.coeff)
wolffd@0:         
wolffd@0:         segments(i).timbre(j) = str2double(...
wolffd@0:             dsegments{i}.timbre{1}.coeff{j}.CONTENT);
wolffd@0:     end
wolffd@0:     
wolffd@0:     % get loudness measurements in dB and time
wolffd@0:     segments(i).loudness = str2double(...
wolffd@0:         dsegments{i}.loudness{1}.dB{1}.CONTENT);
wolffd@0:     
wolffd@0:     segments(i).loudness_time = str2double(...
wolffd@0:         dsegments{i}.loudness{1}.dB{1}.ATTRIBUTE.time);
wolffd@0:     
wolffd@0:     segments(i).loudness_max = str2double(...
wolffd@0:         dsegments{i}.loudness{1}.dB{2}.CONTENT);
wolffd@0:     
wolffd@0:     segments(i).loudness_max_time = str2double(...
wolffd@0:         dsegments{i}.loudness{1}.dB{2}.ATTRIBUTE.time);
wolffd@0: end
wolffd@0: 
wolffd@0: features.segments = segments;
wolffd@0: 
wolffd@0: %% ------------------------------------------------------------------------
wolffd@0: % ---
wolffd@0: % get sections
wolffd@0: % ---
wolffd@0: dsections = data.track{1}.sections{1}.section;
wolffd@0: 
wolffd@0: secstart = zeros(1,numel(dsections));
wolffd@0: secduration = zeros(1,numel(dsections));
wolffd@0: for i = 1:numel(dsections)
wolffd@0:     sections(i).start = str2double(dsections{i}.ATTRIBUTE.start);
wolffd@0:     sections(i).duration = str2double(dsections{i}.ATTRIBUTE.duration);
wolffd@0: end
wolffd@0: 
wolffd@0: features.sections = sections;
wolffd@0: 
wolffd@0: %% ------------------------------------------------------------------------
wolffd@0: % ---
wolffd@0: % get beat and rythm data. the metric data is structured 
wolffd@0: % hierarchically, as each bar contains several beats,
wolffd@0: % which contaisn several tatums. 
wolffd@0: % NOTE: Although the metrum and tempo have been evaluated and fixed
wolffd@0: % on a global scale, the number of bars and tatum vary greatly.
wolffd@0: % ---
wolffd@0: dbars = data.track{1}.meter{1}.bar;
wolffd@0: 
wolffd@0: for i = 1:numel(dbars)
wolffd@0:     
wolffd@0:     % get bar information
wolffd@0:     bars(i).confidence = str2double(dbars{i}.ATTRIBUTE.conf);
wolffd@0:     for j = 1:numel(dbars{i}.beat)
wolffd@0:         
wolffd@0:         % get beat information
wolffd@0:         bars(i).beat(j).confidence = str2double(dbars{i}.beat{j}.ATTRIBUTE.conf);
wolffd@0:         
wolffd@0:         for k = 1:numel(dbars{i}.beat{j}.tatum)
wolffd@0:             
wolffd@0:             % get tatum information
wolffd@0:             if ~isempty(dbars{i}.beat{j}.tatum{k}.ATTRIBUTE)
wolffd@0:                 bars(i).beat(j).tatum(k).time = str2double(dbars{i}.beat{j}.tatum{k}.CONTENT);
wolffd@0:                 bars(i).beat(j).tatum(k).confidence = str2double(dbars{i}.beat{j}.tatum{k}.ATTRIBUTE.conf);
wolffd@0:             
wolffd@0:             else 
wolffd@0:                 % save empty struct
wolffd@0:                 bars(i).beat(j).tatum = struct([]);
wolffd@0:             end
wolffd@0:             
wolffd@0:         end
wolffd@0:     end
wolffd@0: end
wolffd@0: 
wolffd@0: features.bars = bars;
wolffd@0: