Method Of Pitch Feature Extraction And The Application In Phonetic Electric Lock Of DSP

The work of this paper focuses on the method of pitch feature extraction in the speaker recognition system, mainly researching on the performance of the traditional pitch extraction arithmetic. Besides, an innovate algorithm of ICWAF pitch extraction in noisy speech based on the searching and tentative smooth measure is proposed to improve the capability of pitch detection for lower SNR situation. On the basis of the study mentioned above, a speaker real-time recognition system realized with Digital Signal Processing(DSP) is designed and accomplished. Now this system has been used in an automotive phonetic electronic lock.It is just the one and only characteristic, the undeniable characteristic, and the defendable characteristic against false itself that the adoption of biology feature as identity blip will turn into a primary select. The technique focus in terms of identity recognition will also change into such biology feature that takes biology feature as identity verification information. Hence, owing to biological recognition technology, the identity problem in digitization survival will be completely accomplished.Automatic Speaker Recognition is through analyzing and extracting the speaker speech signal, a progress that is to make sure automatically whether the speaker is in the aggregate which this speaker has registered and who is this speaker. Automatic Speaker Recognition has been applied in various domains. In the industry and branch of finance, bond, social insurance, public security, troops, and civil safe attestation .etc, there are broad market demand.The difficulties in speaker recognition focus on feature extraction, so extracting certain feature representing personalities sufficiently will be the hotspot in the future research. In past speaker recognition, the common used features usually are acoustics short-time parameters based on track frequence. Such ones are prone to be interfered by track noise, and don't make use of some long-time or sound inspirator's high layer information in speech. Rhythm is the speaker high layer information and it can depict the rhythm by the envelope and changing contrail of Pitch. The Pitch is key in depicting speaker rhythm, hence extracting pitch exactly is more important. Pitch is the periodic interval caused by vocal cords vibration. Pitch is one of the most important speech peremeters, depicting an important asset of speech inspirator. The estimation of Pitch called Pitch Detection, has been used in various aspects, such as speech recognition, speaker recognition, speech analyse and synthsis, speech coding in low coding ratio, disease diagnosis in pronunciation system, speech guidance of hearing deformity.DSP is a microprocessor suit for disposing digital information in-time. It can collect, transform, filter, estimate, enhance, compress, recognize digital signal. DSP which is a kind of real-time microprocessor in digital signal processoring area has been widely used in automation, military affairs, radar, medical treatment and home appliances, etc. Its advantages are real-time and speedy. When DSP is applied in speech signal processing, it always becomes a good measure in speech recognition, speaker recognition and speech enhancement.The characteristics of contemporary communication is signal digitization, hence, DSP is displaying increasingly important effect in aspect of impelling contemporary information processing digitization, and DSP will bring bigger effect into play with the development of communication technique.It is obvious that the research on the speaker recognition system and the realization of DSP has important theory and application value. The main contribution of this dissertation is listed as follows:ⅠThe pitch detection of the noisy speech signal is introduced. An innovate algorithm of ICWAF pitch extraction in noisy speech based on the searching and tentative smooth measure on the basis of further research in traditional pitch detection methods. The algorithm uses a method of expanded spectral subtraction. At the same time, a method of pitch detection based on ICWAF is given, which improves the robustness and the precision of the pitch detection. Then the detected pitch periods are smoothed based on the algorithm of searching and tentative smooth measure. The experiments show that higher efficiency and good detection capability can be obtained while the SNR equal to -10dB, however the traditional method AWAC cannot achieve such good performance under the same SNR.ⅡThe speaker recognition system is completed and the program is realized by Digital Signal Processor(DSP). In the progress of implement, since the former method didn't consider the correlation between voronio vector and vectors in practice integration, so adopt a new initial codebook design method called sound inspiration; besides, dispose the empty cells in case they impress the quality of codebook. In the process, split the empty cell into a couple of typitcal cells, so it can avoid the possibility new empty cell appears when clustering again; in order to reduce the influence caused by the changing speaker voice, adopt a new method normalizing threshold; add Pitch as a new feature in the speech feature, so the feature depicts corresponding person more sufficiently. Also the debugging results of the programs which adopt different methods are given. After the test in Matlab and DSP, the most efficient one has been selected to apply in the automotive phonetic electronic lock.The main work and content are sum-up as follows:The dissertation consists of six chapters.In Chapter 1, it explains the research contents and principal work of the dissertation. Then we briefly review the development and present situation information of Biometric Identification Technology, pitch feature extraction in speaker recognition and DSP.In Chapter 2, it introduces the basic knowledge of speaker recognition, the traditional method of pitch feature extraction. At the same time, it summarizes the basic information of the soft realization of DSP. All the above contents are the base of later research.In Chapter 3, it researches the traditional pitch feature extraction arithmetic, which consist of two categories: time domain and frequency domain. In time domain, there are Auto Correlation Function(ACF), average Magnitude Difference Function(AMDF), AWAC, PPROC .etc; in the frequency domain, Simplified Inverse Filter Function(SIFT), Cepstrum Function(CEP) and Wavelet Transform Function .etc. The arithmetic in time domain is relatively simple, but since what it disposes is the signal in time domain, such arithmetic prone to be affected by noise. While the arithmetic in frequency domain takes on high extraction precision and good antinoise performance. But it self is more complicated, that can bring some problems in the realization of DSP system. It compares and values these methods, at the meantime gives the corresponding emulation results.In Chapter 4, the pitch detection of the noisy speech signal for lower SNR is introduced. An innovate algorithm of ICWAF pitch extraction in noisy speech based on the searching and tentative smooth measure on the basis of further research in traditional pitch detection methods. The algorithm uses a method of expanded spectral subtraction based on the noise compensation structure ruled by an adaptive decision, which can estimate the noise during speech presence. At the same time, a method of pitch detection based on ICWAF is given, which improves the robustness and the precision of the pitch detection. Then the detected pitch periods are smoothed based on the algorithm of searching and tentative smooth measure. The experiments show that higher efficiency and good detection capability can be obtained while the SNR equal to -10dB, however the traditional method AWAC cannot achieve such good performance under the same SNR.In Chapter 5, the speaker recognition system is completed and the program is realized by Digital Signal Processor(DSP). In the progress of implement, since the former method didn't consider the correlation between voronio vector and vectors in practice integration, so adopt a new initial codebook design method called sound inspiration; besides, dispose the empty cells in case they impress the quality of codebook. In the process, split the empty cell into a couple of typitcal cells, so it can avoid the possibility new empty cell appears when clustering again; in order to reduce the influence caused by the changing speaker voice, adopt a new method normalizing threshold; add Pitch as a new feature in the speech feature, so the feature depicts corresponding person more sufficiently. Also the debugging results of the programs which adopt different methods are given. After the test in Matlab and DSP, the most efficient one has been selected to apply in the automotive phonetic electronic lock.In Chapter 6, general conclusion is set out together with the further research direction and future work.