No abstract

We describe the Lwazi corpus for automatic speech recognition (ASR), a new telephone speech corpus which contains data from the eleven official languages of South Africa. Because of practical constraints, the amount of speech per language is relatively small compared to major corpora in world languages, and we report on our investigation of the stability of the ASR models derived from the corpus. We also report on phoneme distance measures across languages, and describe initial phone recognisers that were developed using this data. We find that a surprisingly small number of speakers (fewer than 50) and around 10 to 20 h of speech per language are sufficient for the purposes of acceptable phone-based recognition.

No abstract

No abstract

Sufficient target language data remains an important factor in the development of automatic speech recognition (ASR) systems. For instance, the substantial improvement in acoustic modelling that deep architectures have recently achieved for well-resourced languages requires vast amounts of speech data. Moreover, the acoustic models in state-of-the-art ASR systems that generalise well across different domains are usually trained on various corpora, not just one or two. Diverse corpora containing hundreds of hours of speech data are not available for resource limited languages. In this paper, we investigate the feasibility of creating additional speech resources for the official languages of South Africa by employing a semi-automatic data harvesting procedure. Factorised time-delay neural network models were used to generate phone-level transcriptions of speech data harvested from different domains.