For reducing requirement of large memory and minimizing computation complexity in a large-vocabulary continuous speech recognition system, speech segmentation plays an important role in speech recognition systems. In this paper, we formulate the speech segmentation as a two-phase problem. Phase 1 (frame labeling) involves labeling frames of speech data. Frames are classified into three types: (1) silence, (2) consonant and (3) vowel according to two segmentation features. In phase 2 (syllabic unit segmentation) we apply the concept of transition states to segment continuous speech data into syllabic units based on the labeled frames. The novel class of hyperrectangular composite neural networks (HRCNNs) is used to cluster frames. The HRCNNs integrate the rule-based approach and neural network paradigms, therefore, this special hybrid system may neutralize the disadvantages of each alternative. The parameters of the trained HRCNNs are utilized to extract both crisp and fuzzy classification rules. In our experiments, a database containing continuous reading-rate Mandarin speech recorded from newscast was utilized to illustrate the performance of the proposed speaker independent speech segmentation system. The effectiveness of the proposed segmentation system is confirmed by the experimental results.

In this paper, we propose a new competitive learning algorithm for training single-layer neural networks to cluster data. The proposed algorithm adopts a new measure based on the idea of "symmetry" so that neurons compete with each other based on the symmetrical distance instead of the Euclidean distance. The detected clusters may be a set of clusters of different geometrical structures. Four data sets are tested to illustrate the effectiveness of our proposed algorithm.

Recently, feature maps have been applied to various problem domains. The success of some of these applications critically depends on whether feature maps are topologically ordered. Several different approaches have been proposed to improve the conventional self-organizing feature map (SOM) algorithm. However, these approaches do not guarantee that a topologically-ordered feature map can be formed at the end of a simulation. Therefore, the trial-and-error procedure still dominates the procedure of forming feature maps. In this paper, we propose a healing mechanism to repair feature maps that are not well topologically ordered. The healed map is then further fine-tuned by the conventional SOM algorithm with a small learning rate and a small-sized neighborhood set so as to improve the accuracy of the map. Two data sets were tested to illustrate the performance of the proposed method.

In this paper, we explore the possibility of applying associative memories for locating frontal views of human faces in complex scenes. An appealing property of the associative-memory-based face detection system is that learning of the associative memory may be achieved by using a simple Hebbian learning rule. In addition, a simple heuristic rule is used to quickly filter a certain amount of nonface images at the very beginning of the whole detection procedure. By using the rule, we won't waste unnecessary computational resources on those nonface images. A database consisting of 74 images was used to test the performance of our associative-memory-based human face detection system.