The Voice Recognition Technology And Application Research

With the wide use of the intelligent apparatus, the speech recognition technologyhas become one of the important means of human-computer interaction. Speechrecognition technology is not only the voice signal processing as an important branchof speech signal processing, but also is an important research topic in the field ofhuman-computer interaction, which used in various instrument, the purpose of whichis to make the machine through the recognition process, the recipient’s voice andtimely response. At present, the speech recognition system is still affected byenvironmental noise, recognition effect, result in the recognition rate is not high.Therefore, how to improve the speech recognition system and reliability has becomethe research hotspot of speech recognition.In this paper, mainly studied the speech recognition algorithms and the keytechnology and the speech recognition system, including speech recognition principle,pretreatment, endpoint detection, feature extraction, pattern matching and speechrecognition system research.(1)For voice recognition pretreatment process of the voice signal acquisitionbased on the noise characteristics, denoising with hard and soft threshold denoisingmethod based on, put forward a kind of improved wavelet threshold method tocomplete the process of removing noise, purify speech purpose, experimental resultsshow that, put forward the improved wavelet threshold method basic to the addition ofnoise, noise caused by burr weakened significantly, to retain the integrity of theimportant information in speech.(2)The pattern matching process identification algorithm characteristics, throughthe analysis of mode matching principle, the traditional DTW algorithm (DTW)characteristics, this paper proposes an improved dynamic time warping algorithm forpattern recognition. The experimental results show that, the improved dynamic timewarping algorithm is compared with the traditional dynamic time warping algorithm,the recognition rate is improved obviously, and the method to use the whole pathconstraints and regional restrictions combining dynamic path search method, reducingthe number of unnecessary calculation, improve recognition speed.(3)To build a speaker-independent continuous speech recognition system ofvocabulary, using TMS320VC5509A DSP development board as the main controller,the main part of a audio collection TLV320AIC23module, power supply module,clock module, JTAG CPLD module and external memory module. The use of C/C++language programming, the improved dynamic time warping algorithm and thetraditional HMM algorithm Viterbi algorithm into the corresponding procedures, inDSP development environment CCS3.3complete system testing and commissioning, validation of two algorithm application. Using MATLAB2010b software and on thepart of algorithm simulation. The experimental results show that, by using theimproved dynamic time warping algorithm recognition rate has reached90.9%, whilethe traditional DTW algorithm recognition rate can reach86.8%, the recognition rateis increased, but the traditional HMM model Viterbi algorithm system identificationrate of89.4%. Therefore, these two kinds of algorithm in the structures of the speechrecognition system can be normal use, and demonstrates the feasibility of speechrecognition systems, the recognition rate reaches the expected target.In this paper, the structures of the speaker-independent continuous speechrecognition system of vocabulary has high use value and extensibility.