摘要： 针对大容量模板库条件下固定音频检索面临检索速度慢的问题，提出一种基于模板子空间的快速固定音频检索方法。利用相同或相似音频数据间存在一定相关性的特点，采用基于模板子空间的方法快速筛选出相似度较高的候选模板集，对候选集内的所有模板进行精细检索以得到最终结果。实验结果表明，与已有方法相比，该方法可以在几乎不损失精度的情况下大幅提高检索速度。

关键词: 固定音频检索, 矢量量化, 子空间, 两阶段法, 部分距离, 感觉加权线性预测

Abstract: According to the problem of slow retrieval in specific audio retrieval under the conditions of large template library, this paper proposes a two-stage specific audio retrieval method based on template-subspace. Utilizing the characteristics of a certain correlation among the same or similar templates, this method uses the method based on template-subspace to choose candidate template set of high similarity, and retrieves all the templates in the candidate set precisely to get the final result. Experimental results show that, compared with existing algorithms, this method can significantly improve the retrieval speed in the case of almost no loss of precision.

Key words: regular audio retrieval, vector quantization, subspace, two-stage method, partial distance, Perceptual Linear Predictive(PLP)

中图分类号: 

• TP391.4