论文部分内容阅读
自动音乐/语音分段是语音识别技术的重要部分。该文采用回声器时频分析计算平均能量谱及定长片段的优化短时低能量比,用Bayes分类器判定类型,并根据内容连续性对分段结果修正;最后采用振幅包络匹配滤波器求所有起始点,对分段结果进一步优化。实验基于多语种电视电台录音和国内电话录音数据展开,结果显示该方法的性能FMeasure可达0.987,较已有分类系统性能有大幅提升,同时处理速度也有大幅度改进。
Automatic music / speech segmentation is an important part of speech recognition technology. In this paper, the average energy spectrum and the optimized short-time low energy ratio of the fixed-length segments are calculated by the time-frequency analysis of the echoes. The Bayes classifier is used to determine the types and the segment results are corrected according to the content continuity. Finally, the amplitude envelope matching filter Find all starting points and further optimize the segmentation results. Experimental results show that the performance of this method can reach 0.987 based on the recording of multi-lingual TV stations and domestic telephone recordings. Compared with the existing systems, the performance of this method has been greatly improved, and the processing speed has also been greatly improved.