Import Audacity Labels to Signal Labeler

This example shows how to import labels created in Audacity™ into Signal Labeler.

Read Labels in MATLAB

You have an audio file consisting of a human voice uttering "Volume up" several times.

Load the audio file to the MATLAB® workspace.

audioFile = "speaker1.ogg";
[x,fs] = audioread(audioFile);
sound(x,fs)

You label the file in Audacity [1] and export the labels to speaker.txt.

Read the labels using readtable. The labels consist of three columns corresponding to the speech utterances along with their respective start and end times (in seconds).

labelFile = "speaker1.txt";
roiTable = readtable(labelFile,Delimiter="tab");
roiTable.Properties.VariableNames = ["StartTime","EndTime","Value"]

roiTable=6×3 table
    StartTime    EndTime        Value    
    _________    _______    _____________

     0.33273     0.74033    {'Volume up'}
      1.2062      1.8716    {'Volume up'}
      2.6785       3.111    {'Volume up'}
      3.7432      4.4087    {'Volume'   }
      4.7331      5.0243    {'Up'       }
      5.4984      5.9809    {'Volume up'}

In order to gain insight into the labels, you plot the audio signal along with a mask corresponding to labeled regions of speech.

Convert the signal regions of interest to a binary mask.

mask=signalMask(table(roiTable{:,1:2},categorical(roiTable{:,3})),SampleRate=fs);
plotsigroi(mask, x, true);

Convert Labels to a Labeled Signal Set

Next, convert the label data to a labeledSignalSet that can be imported to Signal Labeler.

First, define the label type. Specify "roi" (Region of interest) for the label type, and "string" for the label datatype.

labelName = "Speech";
lblDef = signalLabelDefinition(string(labelName),...
                               LabelType="roi",...
                               LabelDataType="string");

Next, create a labeledSignalSet pointing to the labeled audio file. Add the label definition to the labeled signal set.

lss = labeledSignalSet(audioDatastore(audioFile),lblDef);

Set the label values.

roiLimits = [roiTable.StartTime roiTable.EndTime];
setLabelValue(lss,1,labelName,roiLimits,string(roiTable.Value));

Load Labels in Signal Labeler

You are now ready to read these labels into Signal Labeler.

1) Open signalLabeler

2) Click Import, then From Workspace, and import lss.

The audio signal is now available to you along with its labels.

Import a Labeled Dataset

The helper function importLabels creates a labeled signal set corresponding to labels for multiple audio files. In this example, you work with two audio files with labels stored in text files.

Call importLabels to create a signal data set for multiple annotated files. Specify the label name as "Speech".

lss = importLabels("Speech");

You can now load lss in SignalLabeler by following the same steps from the previous section.

function lss = importLabels(labelName)
% IMPORTLABELS Import labels from multiple audio file
%
% The function assumes the labels are stored in *.txt files
% The function assumes the label file name is identical to the 
% audio file name.

labelFiles = dir("*.txt");
labelFiles = {labelFiles.name};
numFiles = length(labelFiles);

% Create an audio datastore pointing to all audio files in the current
% folder
ads = audioDatastore(pwd);

lss = labeledSignalSet(ads);
lblDef = signalLabelDefinition(string(labelName),...
    LabelType="roi",...
    LabelDataType="string");
addLabelDefinitions(lss,lblDef)

for index0=1:numFiles
    filename = labelFiles{index0};
    roiTable = readtable(filename,Delimiter="tab");
    roiTable.Properties.VariableNames = ["StartTime","EndTime","Value"];
    roiLimits = [roiTable.StartTime roiTable.EndTime];
    setLabelValue(lss,index0,labelName,roiLimits,roiTable.Value);
end
end

References

[1] https://www.audacityteam.org/