Research And Implementation Of Audio Fingerprint Algorithm Based On Auditory Mechanism

With the arrival of the era of big data, the amount of in formation on the Internet is growing explosively. The traditional retrieval based on text label can not satisfy the demand of the multimedia retrieval. In recent years, information retrieval based on multimedia content has become a hot research topic. In the content-based multimedia retrieval, the sample retrieval( query-by-example) is convenient, not requiring label information and low requirement for users.For example, people can search the database by submitting a short unknown recording to get the relevant information of the section. Audio fingerprinting is an important form of the sample retrieval, with the advantages of small volume, fast searching, etc.In this paper, we focus on the key issues of the audio fingerprint algorithm, and the main research is as follows:Firstly, aiming at deal with the problem of low recall rate of short query fingerprint, we propose an audio fingerprint generation method based on the auditory mechanism. In the process of selecting a peak point in computing the audio fingerprint, we make full use of auditory masking effect of the human ear, and take peak energy as a criterion to generate masking threshold curve, then the masking threshold curve is used for subsequent selection of peak point. When a new peak point is selected, a superimposed method is used to update the threshold curve.Thus, the peak points which are not easily perceived by human ear are filtered. We propose using the masking effect to build a dynamic threshold curve, and pick out more robust spectrum peak s, the fingerprints which are generated by these spectrum peaks have better robustness and can improve the recall rate of audio fingerprint retrieval. The experimental results show that the method can improve the recall rate obviously.Secondly, according to the problem of large number of audio fingerprinting algorithm parameters, wide numerical range, and it is difficult to find the optimal combination of parameters, Since swarm intelligence can quickly search for optimal solution in high dimensional space, we uses particle swarm optimization algorithm, genetic algorithm on parameter optimization, and in fitness function, we take into comprehensive consideration of the recall rate, accuracy, speed and other key performance indicators. The experimental results show that the two methods can achieve better results, and the performance of particle swarm intelligence is better than that of genetic algorithm.Finally, we use C++ as the development tool and implement the audio fingerprint retrieval system. In this system, an audio fingerprint retrieval module is implemented in the form of dynamic link librar, providing interface functions, enabling rapid retrieval based on fingerprints, robust.