The Research Of Humanoid Robot Speech Recognition Based On HMM Algorithm

Science and technology is increasingly developed, electronic devices willbecom smaller and smaller which has more challenge, which has put forward higherrequirements for the way of human-computer interaction. What to be used to replacethe clumsy keyboard or keys is a big problem that people have been efforts to solve.The convenient, efficient,effective tool to communicate is human language, so voicerecognition technology comes into being and has made rapid development. Thespeech recognition technology is that through the transformation of speechacquisition, feature extraction and modeling the machine turn audio frequency intoa plain text. In recent years, ASR technology now has achieved a lot, which has takethe technology leave for the market from the laboratory, and applied in intelligentrobotic products.This paper is committed to the research of humanoid robot voice recognitiontechnology. Non-specific voice control system has been achieved successfullythrough HMM-based algorithm. Unlike the PC system, humanoid robot voicerecognition has its own limitations, where noise is a major influential factor, and thecurrent HMM algorithm has problems of Local optima, thus how reducing the noiseand improving the HMM algorithm is the main issues to be addressed in this article.Firstly, a thorough research for the basic process for voice recognition has beenmade.Identifying with good ability and noise immunity Mel cepstral are selected forfeature extracting. In order to recognise the voice of more people,we choose to useHMM models. Then I had depth analysis on the method to solve the three classicproblems of HMM,they are Back-forword, Viterbi, and Baum-welch algorithm. Andfor the defects of Baum-welch local optimum and slow convergence, K-meansalgorithm is put forward on the desk which has achieved an improvement. Thisalgorithm is not only to make up for deficiencies Baum-welch algorithm, but alsoimproving the recognition time of the system. For the noise of the laboratoryenvironment and machine noise of the robot itself spectral subtraction algorithm isdesigned to improve recognition rate. Finally humanoid intelligent robot hardwareplatforms and control framework have been introduced, and algorithm been appliedto embedded system From different experimental places,different sound spoken, different gender and different style of acquisition of voice I had a morecomprehensive comparison and analysis for the results,which shows the robustnessand applicability of the system.