| Voice activity detection is the first step of any speech signal processing, and the accuracy of this part has great influence on the subsequent speech analysis, speech coding, speech recognition, and speech synthesis. Accurate voice activity detection can not only greatly reduce the processing time, but also avoid the interference of noise from silence part in speech. So voice activity detection is very important in many speech application fields.The thesis designed a real time voice activity detection system using digital signal processor TMS320VC5416. The characteristics of several speech signal processing algorithms, including short-time energy, averaging short-time amplitude modulation (AM), averaging short-time zero-crossing ratio, short-time auto-correlated function, average magnitude difference function (AMDF) in time domain, and Fourier analyses in frequency domain, were revealed and compared with the aid of Matlab simulation. Based on the results, a simple but practical scheme that combines AM and AMDF for accurate voice activity detection is proposed.To realize real-time detection of speech signal, TMS320VC5416 and TLV320AIC23B made by TI Company were chosen as kernel chips in hardware design. The programs for above scheme were compiled jointly in DSP assembly language and C language. And then the thesis carried out simulation in MATLAB and experiments under CCS environment. Several kinds of input signal, including monochromatic signal, speech, noise and noisy speech signal, were analyzed. During the process the upper and lower limit for speech signal in AMDF, the lower limit for speech signal in AM and the number of front-protect and back-protect frame were properly determined. As a result the overall performance of the system was improved and was capable of detecting voice activity accurately in real-time. Together with its modular structure of the software that ensures easy expansion and transplantation, the system is applicable for many real-time problems. |
PDF Full Text Request