Search

The neural network method is one of the most important methods in the field of speech recognition. In this paper, we propose a new speech recognition method, probabilistic neural network (PNN) ensembles, where the Bagging ensembles method is used to form a speech recognition model with probabilistic neural networks integrated, to implement a speaker-independent English speech recognition system. This paper also demonstrates that before speech recognition, applying segment clustering algorithm to the extracted speech data, i.e., the process of time warping, can ensure the validity of dataset and the performance of PNN. Through experiments, the experimental results show that the PNN ensembles method has faster modeling speed and higher recognition rate than the single BP (Back Propagation) and the BP ensembles method, and has higher recognition rate than the traditional PNN method.

In this paper, the study of methods of multi-parameter objective evaluation of pronunciation is introduced. The accuracy of sentences, the emotional expression, the volume matching degree, tone, speaking rate and rhythm are selected as the parameters for evaluating an English sentence. As the result of the evaluation, an objective rating of the input voice, as well as the feedback are presented to the user. And the experiment shows that the method of this paper is effective.

Approaches of ROI (regions of interest) confirmed and similarity calculated are key techniques for ROIBIR (regions of interest-based image retrieval) system. In this paper, two approaches of confirming ROI and a region-based similarity measure approach are proposed. Two approaches of confirming ROI which consider enough the visibility of image regions,the possibility of designating region and itpsilas weight by users make the image retrieval system include human perception and accord with human vision habits. For the proposed calculating similarity approach, firstly, the similarity of two regions with the combination features, histogram and region shape features, is computed respectively. Then,the similarity of two regions is calculated by the weight product method. The average similarity of all regions is regarded as the similarity between ROI and target image. Experiments show the proposed approaches are available.

A BP neural network-based color space quantization scheme with human perception for the different color during the quantization process is propose in this paper. The method of gathering color samples, the design of gathering software and the method of data processing based on BP neural network for color samples are introduced. This quantization scheme may be used for image color feature extraction and image segmentation based on regions. If this scheme is applied to the image retrieval system, the gap between low-feature and high-level semantic concepts is reduced in a new perspective since it considers human perception into such system with human vision habits.