The Pitch-segmenter Worm

Level
Meter

The level meter also controls the automatic recording transport. Recording starts when the level exceeds the red marker on the left, and stops again when it falls below the black marker on the right. You can drag either marker to set levels appropriate for your environment. Clicking on the green bar sets both start and stop levels simultaneously.

Controls

These control the recording transport.

Start

When the level exceeds the red marker, audio is recorded and appended already available for processing.

Stop

No further audio will be stored.

Clear

Clear the existing recording entirely.

Send

The stored signal is sent to Glasgow CMT for processing.

Above the controls, the status indicator displays the programs current activity.

Audio Window

Displays the audio waveform of the recorded signal. The waveform can be zoomed and scrolled using the [+] and [-] buttons at the bottom of the window, but the whole signal is always sent to the CMT worm server when the sent button is pressed.

Result

Once the results have been processed by the worm server, the results are composed into an animated graphic and returned to the pitch segmenter program to be displayed here.

Instructions for use

1. Download the Java PitchSegmenter GUI from here (138K)
2. Run the program. Mac users can simply double-click the jar file. Linux users can type java -jar <place_you_saved_it>.
3. Use the controls on the  Level Meter  to set the start and stop levels appropriate for your system.
4. Use the  Controls  to start recording, sing into the microphone, and stop the recording. Best results come from a fine legato rendition, in the tenor register or above
5. Press the Send button, and wait. Wait a bit more. The analysis process is quite fast, but assembling the amination can take a few tens of seconds. After a time, an animated worm appears in the.  Result  window. As it moves along the pitch contour, it will change colour from red (no note detected), to blue (note candidate). The final frame shows finalised note candidates after they have been iterated.

This version of the worm is optimised for bad singers (i.e., the author) and will tend to be optimistic when deciding when a note candidate is valid. In practice, with expert performers much stricter limits are applied. The full process is described here.

Nick Bailey,
July 2008.