System and method of Multi model adaptive and voice recognition

摘要

The present invention provides a voice recognition system, the primary voice of the characteristics of each speaker independent speech models and a new multi-model Save by adapting to each, and to ensure a stable multi-model adaptation to speech recognition using a real-time speech recognition is provided. The present invention is a process for selecting either one of the models specified by the speaker, the process of extracting the feature vectors used in speech models in the speaker''s input speech, the pronunciation model and the adaptation information column of feature vectors extracted by using the basic speech model run and stored in the model is specified from among the plurality of speaker models, and the process of setting a flag indicating whether or not to execute adaptation, the process of extracting a feature vector from the speaker voice input for speech recognition, the flag set in the multi-adaptation model the process of selecting only the read is executed by adaptation model, the process of calculating the degree of similarity between the speaker model adaptation values are sequentially compared with the reading of the selected feature vector extracted from the speech input and the flag, the similarity calculation is completed for the entire model When selecting a model having the maximum similarity value comprises the step of executing the speech recognition through the decoding.