Training method and system for mixed speech recognition model of mandarin and Sichuan speech

The embodiment of the invention provides a training method for a mixed speech recognition model of mandarin and Sichuan. The method comprises the steps of determining features of mixed training audiodata, and determining the features as input data for training based on phoneme-based data alignment and Chinese character-based data alignment; inputting to N common intermediate layers, calculating afirst loss function by a first task layer, and calculating a second loss function by a second task layer; training N layers of first task layers based on the first loss function, training N layers ofsecond task layers based on the second loss function, performing multi-task training based on the trained first neural network parameters and the trained second neural network parameters, and training N layers of common intermediate layers. The embodiment of the invention further provides a training system for the mixed speech recognition model of mandarin and Sichuan. According to the embodimentof the invention, phonemes and Chinese characters are used as tasks of multi-task joint training, so that the recognition performance of mandarin and Sichuan talks is improved, and the resource occupation is reduced.