Automatic Audio Topic Detection

Abstract:

Automatic Audio Topic Detection: the development of a device to classify audio documents on 'content' using machine-learning techniques. Given a training corpus of topic-labeled audio, a set of quasi-phonemes ('fenones') is automatically derived through clustering techniques. The documents are then 'fenonically' transcribed, and a common classifier learns to associate fenone transcription features with topic labels. At the same time, a HTK Speech Recognizer is trained on the fenones, so that new audio can be transcribed and labeled by the classifier.

Parking: Visitors requiring a parking pass are asked to contact us at least one working day before the seminar.

Enquiries: sals@ics.mq.edu.au