Main Content

createDatastores

Create datastores pointing to signal and label data

    Description

    example

    [sigdata,lbldata] = createDatastores(lss,lblnames) creates a datastore, sigdata, containing signal member data, and a datastore, lbldata, containing label data from labels specified in the string array lblnames. createDatastores does not apply to sublabels. Set lblnames to one or more parent label names to get the parent labels and the corresponding sublabel values.

    Examples

    collapse all

    Load a labeled signal set containing recordings of whale songs.

    load whales
    lss
    lss = 
      labeledSignalSet with properties:
    
                 Source: {2x1 cell}
             NumMembers: 2
        TimeInformation: "sampleRate"
             SampleRate: 4000
                 Labels: [2x3 table]
            Description: "Characterize wave song regions"
    
     Use labelDefinitionsHierarchy to see a list of labels and sublabels.
     Use setLabelValue to add data to the set.
    
    

    Display the labels for the first member of the set.

    lss.Labels(1,:)
    ans=1×3 table
                     WhaleType    MoanRegions    TrillRegions
                     _________    ___________    ____________
    
        Member{1}      blue       {3x2 table}    {1x3 table} 
    
    

    Get the names of the labels in the set. Create a signal datastore with the signal information and an array datastore with the label information.

    lbls = getLabelNames(lss);
    [sgd,lbd] = createDatastores(lss,lbls)
    sgd = 
      signalDatastore with properties:
    
        MemberNames:{
                    'Member{1}';
                    'Member{2}'
                    }
           Members: {2x1 cell}
          ReadSize: 1
        SampleRate: 4000
    
    
    lbd = 
      ArrayDatastore with properties:
    
                  ReadSize: 1
        IterationDimension: 1
                OutputType: "cell"
    
    

    Display the labels for the first member of the set.

    lbls = read(lbd);
    lbls{:}
    ans=1×3 table
        WhaleType    MoanRegions    TrillRegions
        _________    ___________    ____________
    
          blue       {3x2 table}    {1x3 table} 
    
    

    Specify the path to a set of audio signals included as MAT-files with MATLAB®. Each file contains a signal variable and a sample rate. List the names of the files.

    folder = fullfile(matlabroot,"toolbox","matlab","audiovideo");
    lst = dir(append(folder,"/*.mat"));
    nms = {lst(:).name}'
    nms = 7x1 cell
        {'chirp.mat'   }
        {'gong.mat'    }
        {'handel.mat'  }
        {'laughter.mat'}
        {'mtlb.mat'    }
        {'splat.mat'   }
        {'train.mat'   }
    
    

    Create a signal datastore that points to the specified folder. Set the sample rate variable name to Fs, which is common to all files. Generate a subset of the datastore that excludes the file mtlb.mat. Use the subset datastore as the source for a labeledSignalSet (Signal Processing Toolbox) object.

    sds = signalDatastore(folder,"SampleRateVariableName","Fs");
    sds = subset(sds,~strcmp(nms,"mtlb.mat"));
    lss = labeledSignalSet(sds);

    Create three label definitions to label the signals:

    • Define a logical attribute label that is true for signals that contain human voices.

    • Define a numeric point label that marks the location and amplitude of the maximum of each signal.

    • Define a categorical region-of-interest (ROI) label to pick out nonoverlapping, uniform-length random regions of each signal.

    Add the signal label definitions to the labeled signal set.

    vc = signalLabelDefinition("Voice",'LabelType','attribute', ...
        'LabelDataType','logical','DefaultValue',false);
    mx = signalLabelDefinition("Maximum",'LabelType','point', ...
        'LabelDataType','numeric');
    rs = signalLabelDefinition("RanROI",'LabelType','ROI', ...
        'LabelDataType','categorical','Categories',["ROI" "other"]);
    addLabelDefinitions(lss,[vc mx rs])

    Label the signals:

    • Label 'handel.mat' and 'laughter.mat' as having human voices.

    • Use the islocalmax function to find the maximum of each signal. Label its location and value.

    • Use the randROI function to generate as many regions of length N/10 samples as can fit in a signal of length N given a minimum separation of N/6 samples between regions. Label their locations and assign them to the ROI category.

    When labeling points and regions, convert sample values to time values. Subtract 1 to account for MATLAB® array indexing and divide by the sample rate.

    kj = 1;
    while hasdata(sds)
        
        [sig,info] = read(sds);
        fs = info.SampleRate;
    
        [~,fn] = fileparts(info.FileName);
        if fn=="handel" || fn=="laughter"
            setLabelValue(lss,kj,"Voice",true)
        end
        
        xm = find(islocalmax(sig,'MaxNumExtrema',1));
        setLabelValue(lss,kj,"Maximum",(xm-1)/fs,sig(xm))
    
        N = length(sig);
        rois = randROI(N,round(N/10),round(N/6));
        setLabelValue(lss,kj,"RanROI",(rois-1)/fs,repelem("ROI",size(rois,1)))
    
        kj = kj+1;
        
    end

    Verify that only two signals contain voices.

    countLabelValues(lss,"Voice")
    ans=2×3 table
        Voice    Count    Percent
        _____    _____    _______
    
        false      4      66.667 
        true       2      33.333 
    
    

    Verify that two signals have a maximum amplitude of 1.

    countLabelValues(lss,"Maximum")
    ans=5×4 table
               Maximum            Count    Percent    MemberCount
        ______________________    _____    _______    ___________
    
        0.80000000000000004441      1      16.667          1     
        0.89113331915798421612      1      16.667          1     
        0.94730769230769229505      1      16.667          1     
        1                           2      33.333          2     
        1.0575668990330560071       1      16.667          1     
    
    

    Verify that each signal has four nonoverlapping random regions of interest.

    countLabelValues(lss,"RanROI")
    ans=2×4 table
        RanROI    Count    Percent    MemberCount
        ______    _____    _______    ___________
    
        ROI        24        100           6     
        other       0          0           0     
    
    

    Create two datastores with the data in the labeled signal set:

    • The signalDatastore (Signal Processing Toolbox) object sd contains the signal data.

    • The arrayDatastore object ld contains the labeling information. Specify that you want to include the information corresponding to all the labels you created.

    [sd,ld] = createDatastores(lss,["Voice" "RanROI" "Maximum"]);

    Use the information in the datastores to plot the signals and display their labels.

    • Use a signalMask (Signal Processing Toolbox) object to highlight the regions of interest in blue.

    • Plot yellow lines to mark the locations of the maxima.

    • Add a red axis label to the signals that contain human voices.

    tiledlayout flow
    
    while hasdata(sd)
    
        [sg,nf] = read(sd);
        
        lbls = read(ld);
        
        nexttile
        
        msk = signalMask(lbls{:}.RanROI{:},'SampleRate',nf.SampleRate);    
        plotsigroi(msk,sg)
        colorbar off
        xlabel('')
        
        [X,Y] = meshgrid(lbls{:}.Maximum{:}.Location,ylim);
        hold on
        plot(X,Y,'LineWidth',2,'Color','#EDB120')
        hold off
        
        if lbls{:}.Voice{:}
            ylabel('VOICED','Color','#D95319')
        end
    
    end

    Figure contains 6 axes. Axes 1 contains 4 objects of type line. Axes 2 contains 4 objects of type line. Axes 3 contains 4 objects of type line. Axes 4 contains 4 objects of type line. Axes 5 contains 4 objects of type line. Axes 6 contains 4 objects of type line.

    function roilims = randROI(N,wid,sep)
    
    num = floor((N+sep)/(wid+sep));
    hq = histcounts(randi(num+1,1,N-num*wid-(num-1)*sep),(1:num+2)-1/2);
    roilims = (1 + (0:num-1)*(wid+sep) + cumsum(hq(1:num)))' + [0 wid-1];
    
    end

    Input Arguments

    collapse all

    Labeled signal set, specified as a labeledSignalSet object.

    Example: labeledSignalSet({randn(100,1) randn(10,1)},signalLabelDefinition('female')) specifies a two-member set of random signals containing the attribute 'female'.

    Label names, specified as a character vector, a string scalar, a cell array of character vectors, or a string array.

    Data Types: char | string

    Output Arguments

    collapse all

    Signal data, returned as a signalDatastore (Signal Processing Toolbox) object or an audioDatastore (Audio Toolbox) object.

    Label data, returned as an arrayDatastore object.

    Introduced in R2021a