This script acts as a wrapper around the Trips SpeechIn module, allowing you to record and recognize a monologue without invoking the Trips system. You simply run the below script and the audio_server and speech recognizer are started and record your speech. Segmentation is performed by looking for 1 sec silences in your speaking. The final results are compressed into a tar archive and copied to a temp directory for whoever is running the test.
/u/pmccarty/src/SpeechIn/speechin-collect-data.sh -scenario -sex (m|f)
bash-2.03$ /u/pmccarty/src/SpeechIn/speechin-collect-data.sh -scenario medadvisor Starting Audio Server.. Loading Sphinx.. (this will take up to 30 seconds) ready (say "exit" or enter ctrl-c when done) 1:. 2:. 3:..... 4:... 5:........ 6:.. 7:.data collected. Hit any key to compress and send data... 8834 Killed compressing and sending data.. tar: Removing leading `/' from absolute path names in the archive tmp/pmccarty.200109191409/ tmp/pmccarty.200109191409/speechin.log tmp/pmccarty.200109191409/sphinx.log tmp/pmccarty.200109191409/sphinx.err tmp/pmccarty.200109191409/utt.001.out tmp/pmccarty.200109191409/utt.001.out.lat tmp/pmccarty.200109191409/utt.001.au tmp/pmccarty.200109191409/utt.002.out tmp/pmccarty.200109191409/utt.002.au tmp/pmccarty.200109191409/utt.003.out tmp/pmccarty.200109191409/utt.003.out.lat tmp/pmccarty.200109191409/utt.003.au tmp/pmccarty.200109191409/utt.004.out tmp/pmccarty.200109191409/utt.004.out.lat tmp/pmccarty.200109191409/utt.004.au tmp/pmccarty.200109191409/utt.005.out tmp/pmccarty.200109191409/utt.005.out.lat tmp/pmccarty.200109191409/utt.005.au tmp/pmccarty.200109191409/utt.006.out tmp/pmccarty.200109191409/utt.006.out.lat tmp/pmccarty.200109191409/utt.006.au done. Thank you for your monologue submission for the Trips project.
In the above example, the script was invoked with the medadvisor language model. A total of 7 utteranaces were detected. Including the last utterance which was "exit".
During recognition the system sometimes gets stuck waiting for "exit". If recognition is performing poorly it may never recognize "exit" when you say it. You may have to stop the script with ctrl-c. This will interrupt the recognizer, but will still allow you to send what you have to the researcher.
During recognition, numbered lines will appear with a series of periods. As you speak the periods appear indicating recognition is detecting words. When the cursor advances to the next line, it has detected a 1 sec pause and presumes you have completed your utterance. You may wish to begin a new sentance when you see the cursor return to the beginning of the next line.
CISD Home Page | URCS Home Page