Noise Background Of Isolated Word Speech Recognition Method And Simulation

With the development of modern computer technology, the demands on man-machine communication technologies has increased greatly. Voice-recognition technology appeared on the scene in order to satisfy this requirement. This technology which can recognition humanity's voice accuracy and execute command will be widely used and of important research value.In the past decades of years , voice-recognition technology had made a great improvement in many areas(such as Time ranging from long-Match, establish recognition model, running time, etc). The recognition rate of voice-recognition system has reached a very high standard, especially in a quiet environment. However, the practical applications of calculus voice-recognition system existed many problem which mainly focus on de-noising and accurately-recogniting. In this paper, a voice-recognition system of non-specific people with isolated word in noisy environments is proposed. The research which based on the theoretical of Speech signal, meet a practical applications require of voice-recognition system.This paper is organized as follows: First, paper introduces the processe of voice-recognition system, study voice signal in time-domain, frequency domain and wavelet, analyze preprocessing problems of the voice signal. Preprocessing includes digitized speech signals, pre-emphasis, Sub-frame, plus windows, wavelet denoising and endpoint detection. Among all, paper selective analysis the part of wavelet denoising and endpoint detection. Paper introduces the basic theory of wavelet analysis, the main idea of wavelet threshold de-noising, compares the de-noising results of different threshold and different wavelet under the rules of different threshold rules. Second, the paper compares the characteristic parameters of LPC, LPCC and MFCC. The selected characteristic parameters have a great impact on the real-time and robustness of the entire voice recognition system, so paper analysis and compare the real-time and robustness of those parameters mentioned above.Finally, paper discuss the two kinds voice recognition algorithms: Dynamic Time Warping(DTW) and Hidden Markov Model(HMM), the experiments is carry out ,conducted to recognize the mandarin digitals from 0 to 9 with both algorithms in this system. The iso1ated word recognition systems based on the DTW theory for speaker—independent and speaker dependent are builded while using different feature parameters. The speaker-independent iso1ated word speech recognition system is based on HMM. Experiments shows the system performance effective has improved while the new HMM algorithm is applyed.