基于压缩感知的固定音频检索方法

A Specific Audio Retrival Method Based on Compressive Sensing

固定音频检索是指在待检索音频或音频库中检测和定位与给定样例音频同源的音频片段,它是音频检索中的基本问题之一。针对现有的音频检索方法在特征提取和匹配检索中数据存储空间大、算法复杂和特征信息容易丢失等问题,提出了一种基于压缩感知理论的固定音频检索方法,对具有稀疏性的信号进行采样压缩获得精简的观测值,提高算法的处理效率。首先对经过预处理的音频信号进行感知压缩,得到少量的压缩观测值,对其提取音频特征;然后采用粗检和精检匹配相结合的分层检索方法计算音频在观测值上对应特征的相似度,实现样例音频的精确检索。实验结果表明,该方法在固定音频检索速度和查全率、查准率等指标上具有较优的性能。

Specific audio retrieval which is one of the basic problems of audio retrieval is a means to detect and locate the fixed audio in the given audio or audio library. The existing audio retrieval methods would consume a large data storage space and have a high algorithm complexity,as well as the feature information was easy to be lost,this paper proposes a specific audio retrieval method based on compressive sensing,refers to the characteristics of compressive sensing（ CS） theory in which the sparsely original signal can be compressed sampling to get a small amount of measured value for transmission and processing. First,a small amount of compressed measurements of the original audio signal are gotten by the compressed sampling,and can extract the audio features after the pretreatment. Then,by adopting the method that combines the rough retrieval and the precise retrieval,the audio similarity of the audio features can be calculated. It realizes the accurate retrieval of the specific audio. The experimental results show that this method has a better performance on retrieval speed as well as audio recall and precision,which means that it can retrieval specific audio efficiently and precisely.