Lightweight multi-dialect recognition method and system based on reinforcement learning

The embodiment of the invention relates to the technical field of dialect recognition, in particular to a lightweight multi-dialect recognition method and system based on reinforcement learning. According to the method, on one hand, a first dialect feature belonging to multi-dialect shared information in dialect speech is extracted based on depth separable convolution and M-Transform, then specific information of each dialect is learned based on a specific dialect perceptron, the speech recognition accuracy of a single dialect is improved, and finally, the speech recognition accuracy of the single dialect is improved based on a reinforcement learning means. The weight optimization direction of the model is adjusted to a direction with higher character accuracy and sentence accuracy, so that the accuracy of dialect speech recognition is improved; and on the other hand, feature extraction learning is performed based on lightweight depth separable convolution, M-Transform and a specific dialect perceptron, so th