Parameter optimization by a genetic algorithm for a pitch tracking system

The emergence of multimedia data in databases requires adequate methods for information retrieval. In a music data retrieval system by humming, the first stage is to extract exact pitch periods from a flow of signals. Due to the complexity of speech signals, it is difficult to make a robust and practical pitch tracking system. We adopt genetic algorithm in optimizing the control parameters for note segmentation and pitch determination. We applied the results to HumSearch, a commercialized product, as a pitch tracking engine. Experimental results showed that the proposed engine notably improved the performance of the existing engine in Hum Search.