Research on raw speech isolated word recognition based on Sincnet-CNN model

Abstract

In order to effectively speed up the model training time, reduce the model training parameters and improve the accuracy of raw speech isolated word recognition. An interpretable convolutional filter structure (sincnet) combined with convolutional neural network (CNN) is proposed for the task of raw speech isolated word recognition. On the premise of ensuring the speech recognition rate, the model structure becomes lightweight and the computational complexity is reduced. The experimental results show that compared with the traditional neural network model, the proposed model can effectively improve the performance of raw speech isolated word recognition.