The Emu Segmentation Tool

The Emu segmentation tool is intended to help in the task of cutting a large recording into many small recordings, for example into sentences or words. The tool can make use of a word or prompt list to guide the process and incorporates an automatic, silence based, segmentation algorithm that can be used to generate a first pass set of break points for subsequent manual adjustment. The tool will also optionally write a log file of all operations on the recording, this can be useful in recovering work after a crash or when some incorrect actions have been taken.

Figure 3.8. The main Emu segmenter window

The main Emu segmenter window

The main segmenter interface window displays a section of the waveform being segmented along with the word or prompt list and a set of buttons for navigating the recording and marking cut points. The model of operation of the segmenter is to first build up a list of cut points (the cut list) and then perform the cutting operation, saving the smaller files in a defined location. During the segmentation process you can make notes on particular segments, for example to not that a prompt needs to be re-recorded due to noise or mispronunciation of the prompt words. Both the cut list and the note list can be saved to text files for later reloading or processing by other tools or scripts.

Figure 3.9. The Emu segmenter cut list view.

The Emu segmenter cut list view.

Automatic Segmentation

The segmentation tool includes a simple segmentation algorithm based on RMS and probability of voicing measures from the source signal. This can be used to pre-segment the recording as it is loaded; if this is done then you can navigate through the recording by skipping over successive chunks rather than by moving forward and backward by time.

There are two ways to invoke automatic segmentation: either directly via the Actions menu or using the option (set in the Options menu) to automatically segment files on loading.

Segmentation of files does not produce any visible result but instead stores a list of breakpoints internally which can be navigated by the forward and backward buttons in the interface. The breakpoints found will be saved in a file with the extension .breaks and the same basename as the source file. The tool will offer to reload this file rather than re-running the segmentation algorithm the next time you process this file.

Performing Segmentation

The basic operation of the segmenter is to work through a source audio file and a prompt list locating the start and end of each prompt and saving these to the cutlist.

  • After loading a file and performing automatic segmentation you can navigate through the different segments via the arrow buttons on the interface.

  • If you choose not to use the automatic segmentation algorithm you can navigate through your file in fixed time increments, the default setting is 10 seconds but this can be easily changed via the entry box.

  • If you have a prompt or word list for your recording, load this using the button at the bottom of the window (or the File menu). The words will be displayed in the list at the bottom right of the window. As you add items to the cut list, the selected word will advance by one automatically; you can of course select the appropriate word with the mouse.

  • The output filename which each section of audio will be written to is generated from a prefix (Filename Prefix) and the current word highlighted in the prompt list. For example, if the prefix is set to sp1 and the current highlighted word is artichoke the segment will be saved as sp1artichoke.wav.

  • Once the relevant portion of the signal is displayed you can play it with the button on the left of the waveform and select regions of signal with the mouse (click the mouse and drag to mark the region of interest). Once the correct region for a word has been marked, press the green Cut button to add the region to the cutlist.

  • The start and end time display in the lower half of the interface shows the current start and end of the region of the signal displayed. You can modify these times if you wish to jump to a particular section of the signal.

  • You can note any section of signal for later reference via the Note button. The note list is similar to the cut list but is intended for notes such as the need to re-record a segment. At present the segment is merely marked as having been noted, no text is associated with the note.