论文部分内容阅读
关键音频检测是指从音频库中检索出查询样例,是音频检索的一种重要形式。该文针对传统关键音频检测方法在效率和鲁棒性上的不足分别在预处理、指纹提取以及检索部分进行了优化。在预处理阶段采用基于子带能量比的语音端点检测算法,并在窗函数选择和子带划分方法上进行了改善;在指纹提取阶段采用种子片段选取的方法,并将指纹提取方法改进为子带频谱质心法;在检索阶段通过设定命中次数门限以提高效率。实验结果表明:该文提出的改进系统在查全率、查准率以及抗噪能力提升的同时提高了检索效率,有效地提升了检索性能。
Key audio detection refers to the retrieval of query samples from an audio library, which is an important form of audio retrieval. In this paper, the traditional key audio detection methods in the efficiency and robustness of the lack of pre-processing, fingerprint extraction and retrieval part of the optimization. In the preprocessing stage, the speech endpoint detection algorithm based on the subband energy ratio is adopted, and the window function selection and subband division are improved. In the fingerprint extraction stage, the seed fragment selection method is adopted and the fingerprint extraction method is improved to subband Spectrum centroid method; in the retrieval phase by setting the number of hits threshold to improve efficiency. The experimental results show that the improved system proposed in this paper improves retrieval efficiency while improving recall, precision and anti-noise ability, and improves retrieval performance effectively.