Speech Recognition System Base On The Enveloping Characteristics

This thesis mainly work through the study of speech recognition system, found that speech recognition system can not achieve real-time processing difficulties:the feature extraction of modules, and large amount of calculation complexity algorithm, hardware realization occupied many hardware resources. The traditional speech recognition system is characteristic vector restrictions, low efficiency.First I adopted a new feature extraction algorithm, based on the characteristics of speech signal, the extraction and distract the extremism of speech signal and voice signal corresponding position, extreme of the speech signal feature vector of the envelope. And through the speech signal envelope characteristic of research and analysis, and decrease the system hardware complexity and reduce the computation of the system, increase the envelope characteristic anti-noise ability.Secondly according to the characteristics of the envelope characteristic vector of phonetic system, and puts forward the structure adjustment, joined the voice and judgment module. Speech and judgment in speech envelope characteristic vector module, on the basis of the speech signal waveform, according to two parts were extracted, and their corresponding to the maximum. Through and judgement, can match the scope of speech is reduced, and further to improve the efficiency of the system. Let the policy by hardware and software, the function of speech recognition.Finally, and with linear forecast for the eigenvector of speech recognition system of traditional compared to occupy the hardware resources, reduce 40% of the hardware resources, one word storage space has 4-7Kbit reduced to 3Kbit, through a large number of experimental rate of recognize is 84.5%.