Example evaluating AST

A tutorial example of how to evaluate the Acoustic Signal Typing on a sample of recordings

Download and open the example corpus

Download the TEVA_AST_example.zip file from [To be announced] and extract it to a convenient location. Open TEVA and then click the Open button on the Main page. A file selection window will open. Navigate into the map you just downloaded and select SignaltypeVoiceSamples.Table. Then click Choose. The Main page now contains a display of an empty sound. SignaltypeVoiceSamples.Table contains data on a number of recordings. You must now select the recording you want to use.

Select speakers

Click on the Speaker button (lower left). A new window will open. Click on the = button. You now see a table with 5 columns and 4 lines (enlarge the window if you do not see all of it). The headings read ID, Text, Description, Audio, and AST. Except for the Description column, all columns contain metadata for the recordings mentioned in the Audio column. Close the Info window and click in the TEVA window to view the Speaker window.

The Speaker window has three text fields. The top one contains the ID of the recording, the next field a short description of the speaker (e.g., name or ID, age, sex, etc.). You can change this field, or even create it for a newly recorded voice. If you enter an ID, or just the start of the ID, or row number in the table, the corresponding speaker from the Info window will be read and displayed in TEVA. If you empty the ID field completely and replace it with a single dash -, the current record will be removed from the table.

Put a single 3 in the top field of the Speaker window to select the third entry in the Table. Then Click the Continue button. You now see a display of the recording of the fourth speaker. You see the ID of this speaker and the Pathological Type as it was recorded in the Table.

Find a stable interval

Inspect the recording using the Sound and Spectrogram windows. You can change the frequency range of the Spectrogram by selecting the desired top frequency in the Configuration page. After selecting the most stable part of the speech, decide how the speech can be judged according to following statements as Good, Mediocre, or Bad. See also Acoustic Signal Typing

With these statements in mind, pick an interval where the vowel is most stable. You can use the Zoom in and Zoom out buttons as well as the Previous (move left) and Next (move right) buttons to navigate around.

When you found a part that you think is most stable, select an interval of 200 ms. Click the Select button. You are now asked to click on the position in the display where you want the start of the interval. When you do that, a vertical blue line appears. Now you are asked to click on the position in the display where the end of the interval should come. Click there. Alternatively, you can simply click inside the display window to select the first boundary. After that it proceeds as with the Select button. Make sure that the interval is at least 100 ms long. You can find the times of start and end points at the top of the graph. You can Zoom in or go To selection, but that is not necessary.

Acoustic analysis

Inspect the other displays, Pitch, Ltas, Intensity, and Harmonicity. Below each of these displays, analysis parameters will be printed about the selected interval (see also Analysis and AST categories). Do these acoustic parameters support your evaluation of the preceding statements? Use the Previous (move left) and Next (move right) buttons to move the selected interval around and see how this affects the acoustic parameters.

Entering the Pathological Type

When you have decided what type of pathological voice the recording has, you simply push one of the numbers 1-4, and the type is recorded. The type will be displayed in the window and in the title bar. Push 0 to remove the type.

Speeding up evaluations in long lists

In case you need to examine a large number of recordings in sequence, a simplification exists. Select the Serial button on the Configuration page. When you go back to the Main page, the Speaker button now reads Nxt and has turned blue. Clicking on this Nxt button now automatically reads and displays the next recording in the list. Just click on the Serial button again to reset TEVA to the default behavior.

Saving your work

After evaluating the sounds, the pathological types and remarks added to the Speaker Info table should be saved to disk. Go to the Configuration page and click on Save. Fill in the name you want the file to have (or use the default) and click Save. During the work, every change or addition made to the Speaker Info table is saved into a backup file with the same name as the original file and an added ~ character. When you Save or Close the table, this backup file is removed. Otherwise, it will remain until you open the table again and change something in it. If you accidently forgot to save the table, you can simply open the backup file and save it under the old, or a new, name.

To save the complete analysis of a single recording, use the Print button on the Main page. This will ask for a place and name to save a directory with all the displays and all the analysis data for the current time interval.

Advanced topic: Pre-set local preferences

For some projects, e.g., evaluation of a large corpus by judges or teaching a course, each user should start TEVA using the same preferences settings. For that purpose, the preferences can be pre-set in a file called TEVApreferences.tsv or .tevarc stored in the same directory as the <list>.Table file (the former file is visible, the latter is hidden). When the <list>.Table file is opened, the settings file is read. See the global preferences file for the available settings.

Links to this page


© R.J.J.H. van Son, February 8, 2012