In the existing algorithms for traffic sign recognition, sometimes the training time is short but the recognition rate is low, and other times the recognition rate is high but the training time is long. To resolve these problems, the Convolutional Neural Network (CNN) architecture was optimized by using Batch Normalization (BN) method, Greedy Layer-Wise Pretraining (GLP) method and replacing classifier with Support Vector Machine (SVM), and a new traffic sign recognition algorithm based on optimized CNN architecture was proposed. BN method was used to change the data distribution of the middle layer, and the output data of convolutional layer was normalized to the mean value of 0 and the variance value of 1, thus accelerating the training convergence and reducing the training time. By using the GLP method, the first layer of convolutional network was trained with its parameters preserved when the training was over, then the second layer was also trained with the parameters preserved until all the convolution layers were trained completely. The GLP method can effectively improve the recognition rate of the convolutional network. The SVM classifier only focused on the samples with error classification and no longer processed the correct samples, thus speeding up the training. The experiments were conducted on Germany traffic sign recognition benchmark, the results showed that compared with the traditional CNN, the training time of the new algorithm was reduced by 20.67%, and the recognition rate of the new algorithm reached 98.24%. The experimental results prove that the new algorithm greatly shortens the training time and reached a high recognition rate by optimizing the structure of the traditional CNN.

Articulatory features, which represent the articulatory information, can help prosodic features to improve the performance of tone recognition. In this paper, a set of 19 pronunciation categories was given according to the pronunciation characteristics of initials and finals. Besides, 19 articulatory tandem features, which are the posteriors of speech signal belonging to the 19 pronunciation categories, were obtained by hierarchical multilayer perceptron classifiers. Then these articulatory tandem features, as well as prosodic features, were used for tone modeling. Tone recognition experiments of three kinds of tone models indicate that about 5% absolute increase of accuracy can be achieved when using both articulatory features and prosodic features. When the proposed tone model is integrated into LVSCR (Large Vocabulary Continuous Speech Recognition) system, the character error rate is reduced significantly.

Concerning the changeability of the speech signal caused by co-articulation phenomenon in Chinese speech recognition, a syllable-based acoustic modeling method was proposed. Firstly, context independent syllable-based acoustic models were trained, and the models were initialized by intra-syllable IFs based diphones to solve the problem of training data sparsity. Secondly, the inter-syllable co-articulation effect was captured by incorporating inter-syllable transition models into the recognition system. The experiments conducted on “863-test” dataset show that the relative character error rate is reduced by 12.13%. This proves that syllable-based acoustic model and inter-syllable transition model are effective in solving co-articulation effect.