This project aims to develop an interactive foreign language learning system which will be able to recognise non-native speech. Such a real-time recognition system will be able to perform tasks such as
1) Analysis of student speech (reading a text, describing a picture) in terms of mispronunciations.
2) Pronunciation exercises assisted by visualisations of an estimate of correct speech based on estimating the student's vocal and visualisations of the given speech of the student, thus enabling a comparison. In order to realise such a system the following challenges have to be investigated:
I) Applicability of given speech recognition technology to non-native speech. So far all speech recognition systems are based on native speech and fail when confronted with non-native speech.
2) Derivation of acoustic models to recognise non-native speech and track mispronunciations. Such models will contain statistical mapping procedures from native to non-native language, which could change adaptively depending on fluency levels.
3) Derivation of transfer algorithms to estimate non-native speech given a sample recording of native speech of the same speaker.