Annotated corpus acquisition is a difficult problem in supervised approach. Aiming at the intention recognition task of Chinese spoken language understanding, two weakly supervised training approaches were studied. One is combining active learning with self-training, the other is co-training. A new method of acquiring two independent feature sets as two views for co-training was proposed based on spoken language understanding data in cascade frame. The two feature sets were character features of sentence and semantic class features obtained from key semantic concept recognition task. The experimental results on Chinese spoken language corpus show that the method combining active learning with self-training can minimize manual annotation compared with passive learning and active learning. Furthermore, under the premise of a few initial annotation data, co-training based on two feature sets can make the classification error rate fall in an average of 0.52% with single character feature set.

In three-dimensional sound reproduction with two speakers, Crosstalk Cancellation System (CCS) performance optimization often pay more attention to the effect independently by the factors such as inverse filter parameters design and loudspeaker configuration. A frequency-domain Least-Squares (LS) estimation approximation was proposed to use for the performance optimization. The relationship between these factors and their effect on CCS performance was evaluated systematically. To achieve the tradeoff of computing efficiency and system performance of crosstalk cancellation algorithm, this method obtained the optimization parameters. The effect of crosstalk cancellation was evaluated with Channel Separation (CS) and Performance Error (PE) index, and the simulation results indicate that these parameters can obtain good crosstalk cancellation effect.

For Minimum Phone Error based Maximum A Posteriori (MPE-MAP) adaptation, in order to accurately estimate the center of prior distribution and to improve the recognition performance, the Maximum Mutual Information based MAP (MMI-MAP) adaptation and H-criterion, which was the interpolation of MMI and Maximum Likelihood (ML) criterion, based on MAP (H-MAP) adaptation were used for the estimation of the center of prior distribution, which led to MMI-MAP prior based MPE-MAP (MPE-MMI-MAP) and H-MAP prior based MPE-MAP (MPE-H-MAP). The experimental results of task adaptation show that the two proposed methods both can obtain better recognition performance than MPE-MAP, MMI-MAP and MAP adaptation. MPE-MMI-MAP and MPE-H-MAP can obtain 3.4% and 2.7% relative improvement over MPE-MAP respectively.