This paper proposes an integrated parsing and word sense disambiguation system. The ambiguity problem is solved when introducing semantic knowledge into the parser by modifying the lexical grammar iteratively. Syntactic information is used to deal with polysemous words in the training process. The experimental results show that the new method not only improves the parsing performance, but also has a good performance on word sense disambiguation.option and the closed fuel cycle (CFC) option which consists of the thermal reactor recycle (TRR) and the fast reactor along with thermal reactor recycle (FRR) are calculated. The natural uranium demand, the separate work demand, the nuclear power demand on alternative style of reactors, the nuclear assemblies demand and the disposal demand of nuclear wastes are obtained. According to these results, the FRR option is the optimal strategy with the highest utility of uranium as well as the minimum accumulation of the nuclear wastes.

Aiming at roughly labeled corpora with several hundred hours of speech, a novel approach of constructing text-to-speech system is proposed. This approach realizes automatically cleaning and labeling of large-scale corpora by means of speech recognition, text alignment and syntactic parsing. Furthermore, in order to solve the problems of memory space expansion and time consumption for acoustic model training of large-scale corpora, a fast training method, which can ensure the accuracy of acoustic model, is realized through the optimization of conventional process of model training. Subjective evaluations show that the exploitation of large-scale corpora with rough transcription can achieve significant improvement at 0.5 MOS score in contrast with small-scale corpora with exact transcription.

Three HRTF databases were set up for investigating the relationship between distance and interaural cues (including ITD and IID). The first database is from the calculated spherical head model, the second is the distance-dependent HRTF database for KEMAR manikin, and the third is the distance-dependent HRTF database for KEMAR manikin without pinnae. The results using the three databases confirm that distance play an important role in affecting the interaural cues in proximal region.

Sound source localization and tracking has turned to be one of hotspots in acoustic signal processing area in recent years. It is widely adopted in a lot of applications, such as multimedia conference, intelligent robot, speech enhancement, etc. Adaptive Eigenvalue Deposition Algorithm (AEDA) is one of the effective methods for its robustness performance of noise and reverberation. However, AEDA is suffered from its slowness in tracking variation of time delay of arrival (TDOA) as well as its sensitivity to initial value. Faced with such problems, a Modified Adaptive Eigenvalue Decomposition Algorithm (MAEDA) for time delay estimation is proposed, based on which an emulation system is developed. Experimental results show that the proposed new algorithm works well in sound source location and moving sound source tracking, meanwhile, it overcomes the drawbacks of the traditional AEDA algorithm.

A new statistical prosodic structure model is proposed, which is based on the idea of analyzing and modeling of hierarchical stochastic properties of Chinese mandarin, where three basic levels of prosodic structure are divided as: prosodic word, prosodic phrase, prosodic phrase cluster. Meanwhile, synthesized units selection algorithms, which are suited for large-corpus-based speech synthesis, are described and discussed in this paper. The experimental results show that the proposed model is effective and high performance could be obtained.

The main work of this paper is to apply speech signal processing and speech recognition technologies in speech training for deaf children, and a computer aided speech training system suited for deaf children is designed and implemented. The system is divided into three modules, basic training, articulation training and intelligibility training, which is in the fashion of visual feedback of speech features. Based on the characteristic of deaf childrens speech training and the relation between acoustical feature and physiological feature, the contrast training method and object training method are proposed. The clinical evaluation was carried out in China Rehabilitation Research Center for Deaf Children and got a good result in second and third grade kindergarten. The experimental results show that it is effective for the contrast training method and object training method to correct the deaf childrens voice disorder and articulation disorder.

The analysis of the relative importance of components of MFCC for both speech recognition and speaker recognition using DTW recognizer in various noise environments are given. For English digit and under the Euclidean distance definition, the experiment results show cepstral components from C₂ to C₁₆contain the most useful speaker information, while C₀ and C₁ are usually harm to speaker recognition. Cepstral terms from C₁ to C₁₂ are found to contain the most useful speech information. In both tasks, the additive noise decreases the relative importance of low MFCC terms faster than that of the middle and high MFCC terms, and the decrement depends on the speech SNR. The channel distortion will deteriorate low terms more than the middle and high MFCC terms in both tasks, also.