Workflow

Phase 1 - Initial (automatic) segmentation

  1. Install inaSpeechSegmenter for automatic segmentation.
  2. Run ina_speech_segmenter.py with options -d=smn and -g=false. Results in output.csv.
  3. Convert output.csv to output.txt (format: Audacity label-tracks). An example of an R-script for conversion is added.

Phase 2 - Manual annotation in Audacity

The red numbers in the picture refer to the steps 2-4 below.

Prepare

  1. Import audiofile (MP3) in Audacity: file > import > audio.1
  2. Set view of the audio-track to Spectrogram and select spectrogram scale MEL. Use dropdown menu next in the leftbox of the wave-form: spectrogram > spectrogram settings > scale > MEL.2
  3. Import the label file generated by inaSpeechSegmenter and that has been converted to output.txt to guide manual labelling: file > import > labels.

Tag segments

  1. Add new label track for the relMUSS annotation: tracks > add > labeltrack.
  2. Find the first segment with music or speech that indicates the start of the worship service and label the segments in the new label track accordingly.

Suggestions: use the Speech-segmenter track to select a segment, and press CTRL(CMD)-B to create a new label in the relMUSS track. Drag the borders of the segment to match its required length.

Export label-track to .txt

  1. Remove the Speech Segmenter track to make sure only the relMUSS track with newly annotated segments is visible in Audacity.
  2. Export label-track: file > export > export labels

Audacity-example; part of the relMUSS label-track.

  1. For macOS open Audacity in low resolution; select 'low resulation' in the 'display info' of the application. ↩

  2. In case the audio file consists of two tracks (stereo): use the same dropdown menu and select 'split stereo/mono' and remove one of the tracks from the view. ↩