Development of accent recognition systems for vietnamese speech

The aim of this paper is to present an accent corpus for Vietnamese speech. The corpus consists of 3,000 audio files collected from TV advertisements, shows, interviews and other sources. Subsequently, we create accent speech classification models to compare, analyze and evaluate the advantages and limitations of different approaches such as Random Forests [1], Convolutional Neural Networks [2], ResNet50 models [3]. The experimental results show that the CNN-based model achieves the best performance with the accuracy of 76.1%, 73.9% on the development set and the test set, respectively.