The areas of "mispronunciation detection" (or "accent detection" more specifically) within the speech recognition community are receiving increased attention now. Two application areas, namely language learning and speech recognition adaptation, are largely driving this research interest and are the focal points of this work.;There are a number of Computer Aided Language Learning (CALL) systems with Computer Aided Pronunciation Training (CAPT) techniques that have been developed. In this thesis, a new HMM-based text-dependent mispronunciation system is introduced using text Adaptive Frequency Cepstral Coefficients (AFCCs). It is shown that this system outperforms the conventional HMM method based on Mel Frequency Cepstral Coefficients (MFCCs). In addition, a mispronunciation detection and classification algorithm based on Principle Component Analysis (PCA) is introduced to help language learners identify and correct their pronunciation errors at the word and syllable levels.;To improve speech recognition by adaptation, two projects have been explored. The first one improves name recognition by learning acceptable variations in name pronunciations, as one of the approaches to make grammar-based name recognition adaptive. The second project is accent detection by examining the shifting of fundamental vowels in accented speech. This approach uses both acoustic and phonetic information to detect accents and is shown to be beneficial with accented English. These applications can be integrated into an automated international calling system, to improve recognition of callers' names and speech. It determines the callers' accent based in a short period of speech. Once the type of accents is detected, it switches from the standard speech recognition engine to an accent-adaptive one for better recognition results.
展开▼
