Abstract
This thesis investigates the method for detecting error pronunciation of English vowels in utterances spoken by L2 learners, which requires the techniques from digital signal processing and speech recognition. We propose a text independent approach (which does not require the use of a target utterance) for English vowels error detection and learning. Various studies in formant-based speech synthesis have suggested the importance of formant coefficients; this motivates us to investigate pronunciation assessment using formant information instead of MFCC (Mel-frequency cesptrum coefficients) alone. In particular, we explore the addition of formant information to improve the recognition rates of HMM. Then we propose the use of PCN (pronunciation confusion network) together with a formant-based confidence measure to raise error detection rates. The phonology knowledge about the formant and the articulator is then employing to generate high-level feedbacks to the user. Experimental results demonstrate that automatic generation of reliable pronunciation instruction (without using a target utterance) becomes highly possible.