Service Robot Speech Recognition Research

With the continuous improvement of living standards in recent years,people's demand for home service robot gradually increases.Meanwhile,along with the rapid development of artificial intelligence technology,the request of robot's intelligence continues to enhance too.As an important part of intelligent robots,voice recognition function of robot attracted much attentions.Aim at meeting the special requirements in voice interaction between human and robot in household service,in this paper,specific command words speech recognition method is studied as a key task.First,pretreatment process and the characteristic parameters choosing method of voice signals are introduced,then speaker-dependent and speaker-independent speech recognition systems are constructed respectively by the dynamic time warping technology(DTW)and hidden markov gaussian mixture model(GMM).Short-time zero crossing ratio and short-time energy are combined for speech endpoint detection in this paper.Detection of the biggest mute time,based on double threshold endpoint detection method,makes the accuracy of effective speech segment determination improved significantly,and then effectively enhance the stability of the speech recognition.For the problem of large amount of calculation and slow response of traditional algorithms of dynamic time warping,the maximum threshold processing is set based on the traditional algorithm of dynamic time warping to effectively reduce the amount of calculation and improve the real-time performance of the system.In addition,in order to make the selected template well reflecting the current voice characteristics,the reference template library is constructed based on minimum distance template selection method.Speaker-dependent speech recognition system is established based on the improved DTW technology,and tested in both cases of off-line and real-time.Results show that the system's recognition accuracy of words in speaker-dependent speech can reach 98.3% and 97.3% respectively.The speaker-independent robot voice command recognition system based on GMM model was constructed to solve the problems of model parameters describion,model parameters revaluation and voice state sequence decoding,and the off-line identification experiments results show that the speaker-independent word speech recognition accuracy of the system can reach to 92.7%.