This thesis established the DSP-based and PC-based system for speech keyword retrieval and recognition according to the same basic algorithm. This system does not need to train speech models, and the keywords and describing sentences do not put the limit of the number of words and could be any language. Before calculating the speech features, the speech signal need to be pre-processed. The pre-process includes DC bias removing, segment, Rabiner & Sambur end point detection, pre-emphasis, and windowing. About the speech features, the system used 12 degrees of Mel-Frequency cepstral coefficient and 12 degrees of delta coefficient to make a 24-degreed speech feature. The key point of the system is the process of pattern comparison. The system adopted dynamic time warping cooperating with one pass algorithm to improve the optimal process. In order to attain the DSP system, using an optimum likelihood ratio threshold to be the determine standard for not keyword rejection. All of the keywords use the same threshold in the method. It improves the original method which uses least differential to set up the threshold by reducing the requirement of ram. After testing in the experiments, the speech keyword retrieval and recognition system both have great recognition and efficiency.

In this thesis, the PC base and DSP base speech keyword retrieval and recognition systems could work. The keywords and describing sentences will not have the limit of word length and could be any languages. Besides, training speech models is not needed anymore. It means that the database gets its expansibility without training speech models again. We can establish the system on the PC base, and calculate the program with fixed-point DSP board. In the processing of speech signal, lots of mathematical functions will be required. We must reach its immediately effect, so that the system could be useful. In addition, compared with floating point, the fixed point DSP cost much less; it makes the system nearer to users. After being tested by experiments, the speech keyword retrieval and recognition system got great recognition and efficiency.

abstract: In recent years, conventional convolutional neural network (CNN) has achieved outstanding performance in image and speech processing applications. Unfortunately, the pooling operation in CNN ignores important spatial information which is an important attribute in many applications. The recently proposed capsule network retains spatial information and improves the capabilities of traditional CNN. It uses capsules to describe features in multiple dimensions and dynamic routing to increase the statistical stability of the network. In this work, we first use capsule network for overlapping digit recognition problem. We evaluate the performance of the network with respect to recognition accuracy, convergence and training time per epoch. We show that capsule network achieves higher accuracy when training set size is small. When training set size is larger, capsule network and conventional CNN have comparable recognition accuracy. The training time per epoch for capsule network is longer than conventional CNN because of the dynamic routing algorithm. An analysis of the GPU timing shows that adjusting the capsule structure can help decrease the time complexity of the dynamic routing algorithm significantly. Next, we design a capsule network for speech recognition, specifically, overlapping word recognition. We use both capsule network and conventional CNN to recognize 2 overlapping words in speech files created from 5 word classes. We show that capsule network achieves a considerably higher recognition accuracy (96.92%) compared to conventional CNN (85.19%). Our results show that capsule network recognizes overlapping word by recognizing each individual word in the speech. We also verify the scalability of capsule network by increasing the number of word classes from 5 to 10. Capsule network still shows a high recognition accuracy of 95.42% in case of 10 words while the accuracy of conventional CNN decreases sharply to 73.18%. / Dissertation/Thesis / Masters Thesis Electrical Engineering 2018

This thesis surveys theoretical background for automatic keyword suggestion system. It contains overview of current statistical term recognition methods and methods for evaluation of automatic term recognition systems. Based on the known approach the thesis specifies possible enhancements. It explores unifying keywords using thesauri, input text filtering and correction of word forms.