Resource development and experiments in automatic SA broadcast news transcription

Abstract

We present a description of the development and evaluation of a first South African broadcast news transcription system. We describe a number of speech resources which have been collected in the resource-scarce South African environment for system development purposes: a 20 hour corpus of South African English (SAE) broadcast news, a 109M word corpus of South African newspaper text collected for language modelling purposes, and a 60k word SAE pronunciation dictionary. The development of our system is based on similar state-of-the-art broadcast news transcription systems and uses cross-word triphone HMMs, MF-PLP features and per segment cepstral mean and per-bulletin cepstral variance normalisation. Our final system achieves a word error rate of 24.6%. We find that reasonable performance is achieved on spontaneous and telephone speech in our test data. Finally, we consider the recognition of MP3-compressed audio and show that performance deteriorates only at low bit-rates.