Audio Fingerprinting Technology And Its Application In The Copyright Of Broadcast’s Music

With the rapid development of internet radio, people can often see that musical works being used illegally, which seriously infringed the copyright of the owners, and strike the people’s enthusiasm of a creation. Audio fingerprinting technology has an extremely wide range of applications in music retrieval, and is robust to the distortion. Therefore, monitoring of network music’s copyright research has an important significance for radio stations.In this thesis, a system based on the audio fingerprinting extraction algorithm and silence detection algorithm is designed and implemented, which can monitor the music copyright of broadcast. In the part of music recognition, this paper proposes a multi-resolution analysis based audio fingerprint extraction algorithm, and in the part of preprocessing, a robust silence detection algorithm for radio stream is proposed. specific aspects of innovations and works have been done as the following:1. Aiming to the problem that there may occur pitch shifting during transmission, in this paper, a robust Audio fingerprint extraction algorithm which is based on Multi-Resolution Analysis(MRA) is proposed. The algorithm makes a MRA on the input audio, and takes out the extreme point of spectrum. After the combination of every two extreme points, the point pairs forms, and then a audio fingerprint model on the time and frequency is established by the point pairs. In order to make a MRA on the input signal, an IIR filter bank is used after a FFT on the signal. Thus, the frequency axis of spectrum is exponentially divided, which is similar to the division of musical scale. Therefore, MRA is particularly suitable for the analysis of the music signal, and as proven, the proposed audio fingerprint algorithm is not only robust to pitch shifting, but also have a nice real-time characteristics.2. Taking into account the continuity of broadcast audio, this paper proposes a silence detection algorithm, then the continuous audio stream is segmented into relatively independent pieces of music content. In the thesis, we firstly use every one of short-time energy, short-time zero crossing rate and spectrum flatness measure to deal with the silence detection, and analysis its performance. Then, making use of the above three characteristics to determinate whether the audio frame is music or silence(also means background noise). Finally, the silence detection is implemented. The experiments prove that the algorithm is guaranteed under the premise of real-time processing audio streams to obtain a higher endpoint detection accuracy. Besides, the silence detection algorithm has strong robustness to noise.3. Based on the study on audio fingerprint technology and silence detection algorithm, this paper designs and implements a system which can monitor the music copyright of broadcast. This system firstly uses the proposed silence detection algorithm to segmented the internet radio stream, and the proposed audio fingerprint extraction algorithm is used to get the audio fingerprint of radio stream; After that, hash mapping algorithm is used to retrieval the sub-fingerprint; At last, the broadcast’s music copyright monitoring system is finished. The experimental results show that the system achieves a nice performance on broadcast monitoring.