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.

This paper describes the development of text-to-speech corpora for four South African languages. The approach followed investigated the possibility of using low-cost methods including informal recording environments and untrained volunteer speakers. This objective and the additional future goal of expanding the corpus to increase coverage of South Africa's 11 official languages necessitated experimenting with multi-speaker and code-switched data. The process and relevant observations are detailed throughout. The latest version of the corpora are available for download under an open-source licence and will likely see further development and refinement in future.

Introducing pronunciation variants into a lexicon is a balancing act: incorporating necessary variants can improve automatic speech recognition (ASR) and spoken term detection (STD) performance by capturing some of the variability that occurs naturally; introducing superfluous variants can lead to increased confusability and a decrease in performance. We experiment with two very different grapheme-to-phoneme variant prediction techniques and analyze the variants generated, as well as their effect when used within fairly standard ASR and STD systems with unweighted lexicons. Specifically, we compare the variants generated by joint sequence models, which use probabilistic information to generate as many or as few variants as required, with a more discrete approach: the use of pseudophonemes within the default-and-refine algorithm. We evaluate results using three of the 2013 Babel evaluation languages with quite different variant characteristics -Tagalog, Pashto and Turkish -and find that there are clear trends in how the number and type of variants influence performance, and that the implications for lexicon creation for ASR and STD are different.

We present the design and development of a South African directory enquiries (DE) corpus. It contains audio and orthographic transcriptions of a wide range of South African names produced by firstlanguage speakers of four languages, namely Afrikaans, English, isiZulu and Sesotho. Useful as a resource to understand the effect of name language and speaker language on pronunciation, this is the first corpus to also aim to identify the "intended language": an implicit assumption with regard to word origin made by the speaker of the name. We describe the design, collection, annotation, and verification of the corpus. This includes an analysis of the algorithms used to tag the corpus with meta information that may be beneficial to pronunciation modelling tasks.