Research On Music Retrieval Based On Content

In the music retrieval system of query by humming, the user only need to hum part of a song and can get the song he or she wanted. In the pattern recognition field, musical information retrieval (MIR) is receiving more and more attention, there are some difficulties in the query by humming fields, such as pitch detection,feature extraction, music melody matching, environment noise and et al.Some researches about content-based musical retrieval technique are carried out as follows:(1) Feature exaction: this paper analyzed audio signal on the time domain characteristic and the frequency domain range characteristic, different features combination can be applied in different audio retrieval.(2) Endpoint detection: this paper introduced traditional method of endpoint detection, proposed a algorithm of endpoint detection based on Bayesian decision, realized endpoint detection based on energy tracking and zero-crossings counts after center clipping, contrasted the three methods through experiment, confirmed this algorithm is effective.(3) Pitch detection: this paper introduced traditional method of pitch detection, realized pitch detection on voiced by autocorrelation function.(4) Melody-matching algorithms: this paper enumerated melody-matching algorithms by music data form classification, introduced the DTW algorithm in detail, and realized melody matching by this algorithm.Feature exaction is more accurate after endpoint detection, Improved DTW algorithm's recognition effect is increasing after endpoint detection.