Extending Gong for the Support of Teaching Speech Recognition

Project Title

Extending Gong for the Support of Teaching Speech Recognition

Project Leader

Prof Brian Mak

School / Dept

SENG / COMP

Project Duration

Jan 2005 - Dec 2005

Project Description

This project intended to enhance the GONG system to support the tuition of speech recognition. Software available from Cambridge University was used to extend the system so that speech segmentation techniques might be effectively taught to students. Student feedback and observation during usage was used to refine the development.

Project Outcome

• In this adaptation project, the original Gong system was enhanced to support teaching speech recognition. Specifically, a speech recognizer wholly developed by our own speech research group was integrated into the Gong system. Different acoustic models, grammar networks, and dictionaries could also be loaded into the system so that they could be compared in terms of their capabilities to produce better recognition accuracies. Their differences could be visually compared by performing word-by-word “forced alignment” on a common recorded message using the common recognizer. In addition, the recorded message might also be visualized in the time domain as a waveform, or in the spectral domain as a spectrogram.

• The new Gong system was used by the students of the COMP621F course offered in Spring 2005, and a follow-up survey showed that most students agreed that the new Gong was useful in comparing different acoustic models, and helped visualize input speech message.

• In brief, a major deliverable was a new Gong system that has a built-in speech recognition capability. It was believed that there are many teaching applications for the new Gong, and it could be used to enhance the learning of speech recognition technologies.

Status

Completed

Project Documents
(Only accessible by HKUST users)

• Project Plan
• Final Report

Adaptation

Adaptation from Gong, a web based communication tool