MAHIR System: Unsupervised Segmentation for Malay Spoken Broadcast News Stories

Abstract

Current studies on spoken document retrieval (SDR) systems concentrate on building strong systems using an approach that reduces the impact of automatic speech recognition (ASR) on retrieval performance. Herein we tend to propose the SDR system, the main goal of that is to reduce the effect of ASR transcription errors on retrieval performance. An automatic speech recognition system is employed to convert the Malay spoken broadcast news to text. The performance of unsupervised learning is evaluated on the Malay broadcast news using apriori algorithm.