|
ASR (automatic speech recognition) allows people to interact with the Lwazi information system by voice. The speaker-independent ASR system is able to recognise carefully-spoken key words in any of the 11 official languages of South Africa.
The HLT team developed HEasy, a new speech recognition toolkit that builds on HTK (Cambridge University), and that provides an easy-to-use interface for building language and acoustic models - the main components of a speech recognition system.
|
|
Speect systems automatically transform digital text into speech, allowing the information provided on the Lwazi telephone-based system to come from dynamic and frequently-updated sources. In 2004, the HLT Research Group developed the first isiZulu TTS system, which was also the first open source TTS system for any indigenous South African language. For the Lwazi project, the HLT group took on the daunting task of developing Speect for each of the 11 official languages in South Africa. To this end, the Group developed the Speect toolkit, which enables the creation of concatenating speech synthesis systems in new languages.
 Download Speect
|
|
An open source software PBX (private branch exchange), is used as the underlying telephony platform for the Lwazi telephone service. The current focus is the integration of the TTS and ASR technologies with the Asterisk telephony platform. An application development framework is also being created to facilitate the generation of voice-based applications without needing to know the details of the underlying telephony system.
|
