The Method Based on Spectrogram's Image Classification for Chinese Dialect's Tone Recognition

Abstract

In China, linguists are facing the dilemma of the rapid disappearance of dialects. The protection and research of dialects are of great significance to Chinese regional culture, history, nationality and language. At present, the tones of Chinese dialects are recognized by the professional trained researcher's human ear. Although some speech signal processing techniques are also used to observe the fundamental frequency, spectrogram, frequency, amplitude and other characteristics of speech signals, these data cannot be used as the description standard of tone characteristics. From another point of view, this paper uses the spectrogram, which represents the comprehensive characters of speech, to identify dialect tones. The characters classification of speech signals is transformed into the feature classification of images. A simple convolutional neural network model is proposed for tone recognition of dialects, and a good recognition rate is achieved with a small amount of data.