The Design And Implementation Of Radio Music Retrieval System

The internet radio is an emerging application under the development of mobile internet technology. Users always hope to know details of the song being broadcasted while listening to the radio. But the reality is that few of the popular internet radio softwares provide this function. Based on this kind of demand, the internet radio application needs to add music retrieval function to send back information of the song currently playing timely.According to the functional requirements mentioned above, and combining with the improvements to the existing music retrieval algorithm, we specifically design and implement a music retrieval system for the internet radio in this thesis. The system can receive, parse and retrieve the broadcasting stream, then upload the retrieved song information to the radio server for release successfully. The specific content and innovation work is as follows:1. Since the radio broadcast may be affected by the noise and speed changes, the existing fingerprint extraction algorithms always can’t meet the performance requirements, so in this thesis, we apply a extraction algorithm based on the time- frequency location and energy characteristic information. Firstly, the proposed algorithm uses the Mel scale filters to divide the spectral frequency to subbands; then locates the key-points with strong noise resistance and high credibility according to the improved key-point extraction algorithm; and finally codes the fingerprint with both the time- frequency locations of the key-points and the energy variation tendencies nearby. The experiment results show that the algorithm improves the robustness performance to the noise and speed changes.2. The proposed algorithm uses hash map to store the fingerprint metadata. In the fingerprint query step, we apply an optimized query algorithm based on double filter ing matching. According to the metadata of the candidates, the double filtering matching algorithm improves the query speed and accuracy. In order to reduce the influence of random start of the query music segments, the proposed algorithm divides multiple sequences from interval positions of the frame. At the same time, in the double filtering matching process, we optimize the combination values of the time offset threshold and hit threshold, it improves the retrieval accuracy.3. Based on the improved music retrieval algorithm and system functional requirements, we design and implement the internet radio music retrieval system. We construct the music fingerprint database and metadata database with the improved music fingerprint extraction algorithm and the hash map storage mode. With the class module partition and multi- threaded control, the system can accomplish the multi-radio’s simultaneous and continuous retrieval structure. The retrieval process module applies the FF mpeg codec library to receive and parse the broadcast stream, the double filtering matching algorithm to retrieve music information and the http protocol to upload retrieval results. The experiment results show that the system can accomplish the multi-radio music retrieval function reliably and stabily.