关键词: 有声出版物, 语音分割, 时间阈值自适应, 相似性分析

Abstract: In order to achieve the automatic speech segmentation of audio publication, the system model with adaptive time threshold adjustment and similarity detection was established. The adaptive threshold adjustment algorithm was put forward to set the time threshold based on the prior knowledge of script. In order to improve the anti-interference and adaptive abilities, a new algorithm based on the similarity detection of speech units was provided. The test results show that the accuracy of segmentation is 100% when the system has no interference, but the accuracy of segmentation is 98.8% when two interference signals are added to each speech files on average. Practical application shows that the system can meet the automatic segmentation requirements.

Key words: audio publication, speech segmentation, adaptive threshold adjustment, similarity analysis