Acoustic Distinguishing Of Easily Confused Mandarin Phone

As the integration of world economy and the enhancement of our country's position in world economy, learning Chinese has been the urgent requirement for more and more representative people at home and abroad, because of the sharply increase of learner, Chinese teaching resources go short. Computer-assisted language learning can save Chinese teacher resources on the one hand, at the same time it can also meet the learners'requirements to learn Chinese independently at anytime and anywhere, which is an effective method to solve the problem.Nowadays the main technology of Computer-assisted Standard Mandarin learning is making use of speech recognition frame, the disadvantage of this frame is that the acoustic model has a limit discrimination ability of similar pronunciation. To make up the disadvantage of speech recognition frame, This paper discusses in detail the discrimination of easily confused mandarin phone pairs in the Computer-assisted teaching, directs towards acoustic feature of these phone pairs, three new feature extraction algorithms are proposed, Gaussian Mixture Modeling (GMM) is used to classify the new features. A new recognition algorithm of easily confused phone pairs is proposed.In the discrimination frame of easily confused phone, according to the flat tongue and raised tongue consonant with the same articulation manner and different articulation places, the paper proposes the method which adopts the energy concentrated frequency area as the divisional character. Directing towards the different energy distributing of flat tongue and raised tongue with Mel triangle filters extracting energy sum of high and low frequency, the method based on Mel Frequency Cepstrum Coefficient (MFCC) feature is proposed. Aiming at the human ears auditory model, by using the wavelet packet decomposition, describing frequency spectrum of easily confused phone pairs more accurately, the method based on Perceptual Linear Predictive (PLP) feature is proposed. The paper uses GMM to classify the new feature for phone discrimination. The experiment is done by different order of GMM, which attains the purpose of recognition the phone pairs.