As an important carrier of information, since sound can be collected quickly and is not limited by angle and light, it is often used to assist in understanding the environment and creating information. Voice signal recognition technology is a typical speech recognition application. This article focuses on the voice signal recognition technology around various deep learning models. By using deep learning neural networks with different structures and different types, information and representations related to the recognition of sound signal samples can be obtained, so as to further improve the detection accuracy of the sound signal recognition detection system. Based on this, this paper proposes an enhanced deep learning model of multi-scale neural convolutional network and uses it to recognize sound signals. The CCCP layer is used to reduce the dimensionality of the underlying feature map, so that the units captured in the network will eventually have internal features in each layer, thereby retaining the feature information to the maximum extent, which will form a convolutional multi-scale model in network deep learning Neurons. Finally, the article discusses the related issues of Japanese speech recognition on this basis. This article first uses the font (gra-phonem), that is, all these Japanese kana and common Chinese characters, using a total of 2795 units for modeling. There is a big gap between the experiment and the (BiLSTM-HMM) system. In addition, when Japanese speech is known, it is incorporated into the end-to-end recognition system to improve the performance of the Japanese speech recognition system. Based on the above-mentioned deep learning and sound signal analysis experiments and principles, the final effect obtained is better than the main effect of the Japanese speech recognition system of the latent Markov model and the long-short memory network, thus promoting its development.

No abstract