Dissertation > Excellent graduate degree dissertation topics show

Parallel Optimization Method in Language Model for Mandarin Speech Recognition

Author: JinZuo
Tutor: SunJiaSong
School: Tsinghua University
Course: Information and Communication Engineering
Keywords: Speech recognition N- tone pinyin grammar model Parallel Optimization Throughput
CLC: TN912.34
Type: Master's thesis
Year: 2010
Downloads: 87
Quote: 0
Read: Download Dissertation

Abstract


The language model is an important part of the speech recognition system , with the increased complexity of handling voice and corpus volume becomes large , the position of the rear end of the identification system , as the language models in the system is indispensable . Select the appropriate language model , and can greatly improve system performance , handling more natural speech recognition results closer to the speaker wants to express the true meaning . With our research , we found that in addition to the above indicators , the speech recognition system as a dialogue system that should provide to the user a friendly interface , higher recognition speed available to users . Therefore, the demand for the throughput of the system is gradually revealed. At the same time , the multi-core technology has been the rise of technology and parallel , which allows us to see the new space to improve system performance . Make good use of parallel technology , its application to the appropriate place of the system can exponentially enhance system performance . In this context , we use these technologies to optimize the language model of the existing Chinese tone recognition system , a new data model Trigram model quickly find and cache optimization based on the N- language model . New improvements proposed method can improve the throughput of nearly 3 times , preferably improve the performance of the system . In addition, in order to adapt to configure lower environmental configuration , we also proposed to optimize the language model in the calculation of high memory and CPU dependent , such as built-in functions , using the look-up table method to optimize the program consumes height calculations ( such as log calculation) , the removal of redundant computation program . These methods also improve the performance of the system can be small amplitude . After that , we use the parallel optimization techniques language model in the forward-backward algorithm parallel , so that the data in a shorter time can be adequately addressed improve the human-computer interaction , the waiting time and improve the real-time rate . The paper describes the principle of the different lengths of the parallel systems of a candidate sequence , and gives the realization . Finally, the experimental results on the 120 voice when 644.3 seconds long test data : real - time only 0.2375 , retrieval speed has been greatly improved . Speedup of 3.3 or more , indicating the superiority of the system .

Related Dissertations

  1. Multiple ANN/HMM Hybrid Used in Speech Recognition,TN912.34
  2. The Design of a DSP-Based Robot Speech Command Recgnition System,TN912.34
  3. The Design and Research of Health Management Based on Smartphone Environment,TN929.53
  4. Discovery and Mechanism Study of Mitochondrial Mild Uncoupler,R341
  5. The Discovery and Biological Activity Study of ER Stress Modulator,R96
  6. Remote sensing data processing grid platform design and initial implementation,TP79
  7. Screening and Identification of Zearalenone-degrading Strain and Optimization of Zen-degrading Conditions,TS201.3
  8. The Implementation of TCP Performance in Mulit-hop Ad Hoc Networks,TN929.5
  9. The Implementation and Performance Analysis of IEEE802.11 Rate Adaptation Algorithms,TN929.5
  10. Research and Improvement on TCP Performance in Heterogeneous Wireless Networks,TN929.5
  11. Research on Dynamic Resource Allocation Strategies for TD-LTE System,TN929.5
  12. Research and Design of FFT Processor for MIMO-OFDM Applications,TN919.3
  13. Mobile audio and video interactive platform of business execution,TN915.09
  14. High Throughput XTS-AES encryption algorithm for hardware implementation,TP309.7
  15. Shenzhen Telecom CDMA wireless network planning studies and recommendations,TN929.533
  16. Wireless networks based on network coding opportunity Routing,TN929.5
  17. Underwater acoustic communication link based on adaptive HARQ technology research,TN929.3
  18. Wireless environment based on the best listener of the network size range,TN92
  19. Study on the Demonstration Method of New Regional Airport for Navigation Feasibility,F562
  20. Research on Hmm-based Speech Recognition System of the Robot,TN912.34
  21. MFCC -based speech recognition system to improve research and design,TN912.34

CLC: > Industrial Technology > Radio electronics, telecommunications technology > Communicate > Electro-acoustic technology and speech signal processing > Speech Signal Processing > Speech Recognition and equipment
© 2012 www.DissertationTopic.Net  Mobile