In this thesis, we have investigated the feasibility of Computer-Assisted Pronunciation Learning system being developed to foster improvement in L2 (second language) pronunciation of non-native learners using speech recognition technology. To avoid the imperfection of speech recognition, we adopted stable and reliable techniques: phonemic segmentation under a given transcription and the classification on the place and manner of articulation. The use of articulatory features on articulation is advantageous especially for L2 pronunciation learning because they are directly linked to the articulation configuration, so that it is possible to correct the learner's pronunciation problems. Each module successfully integrated signal processing, acoustic analysis, segmentation and scoring, feedback instruction, speech rhythm training, and user-friendly graphic interfaces. The system models were systematically well designed to achieve our research goal, that is to say, to develop native-like pronunciation of non-native learners by machine.
From our experiments, we reached the following primary conclusions:
into the vowel diagrams was
adopted in our CAPL system. For consonants, the classification based
on the place and manner of articulation, which is our first trial in
CALL field, demonstrated its effectiveness to improve the
pronunciation errors that result from linguistic gaps between L1 and
L2.
Observation as part of a methodology for evaluating CALL system is especially important at an early stage in the development of new ideas into tangible systems, firstly in order to ensure whether other existing other system can be instead used as effectively as possible, and secondly to provide guidelines for the production of an improved CALL system. In general, the observation has been divided into three major aspects. A preliminary evaluation was completed by the linguistic expert, language teachers, and the opinions of non-native learners.
