Speaker Adaptation Techniques Research For Traffic Broadcast Audio Information Retrieval

In recent years, artificial intelligence began to appear more and more of them indifferent scenes, Unmanned vehicles traffic platform broadcast audio informationretrieval systems is one of the applications. Broadcast audio traffic informationretrieval system based on speech recognition, natural language processing technologyplatform for auxiliary unmanned vehicle intelligent route planning system. In thissystem, speech recognition technology is one of the key technologies. At present, thespeaker-dependent speech recognition system has reached a satisfactory level, buttraffic radio speaker variability factor present in the audio information retrieval lead toperformance recognition system has seen a dramatic decline. So traffic audioinformation retrieval system to move towards practical difficulties must be overcomespeaker variation.In this paper, the effects of different speaker systems for identifying, analyzing theacoustic differences between the speakers and discussed a variety of speakeradaptation technology, two typical adaptive algorithms: maximum likelihood linearregression (Maximum Likelihood Linear Regression, MLLR) and the maximumposterior probability (Maximum a Posteriori, MAP). Herein also attempts to combinethe two algorithms applied, complement each other, the recognition rate comparedwith two improved algorithms alone. Then it was also studied in two adaptivealgorithms in the feature space: Constraint Maximum Likelihood Linear Regression(CMLLR) and Feature Space Gaussianization (FSG). CMLLR role in the algorithm isa feature space of the linear transformation method, and the principle of the adaptivealgorithm is the same as MLLR; FSG algorithm which is applied to a method ofnon-linear transformation of the feature space, and it is a completely unsupervisedmethods. Finally, we study the effects of traffic audio information retrieval system forits adaptive algorithm, and to which the adaptive algorithm is applied to the system,making the performance of the system has been greatly improved.