Fault-Tolerant Music Search by New Ranking Order Algorithm

Music information retrieval is an active area of research with high practical relevance. Humming a melody is one natural way to overcome the input restrictions when searching for music. In this presentation we concentrate on a new ranking order algorithm to match melody input to a music database containing polyphonic music as sequences of notes. The new algorithm achieves high melody recognition rates and shows graceful degradation in the presence of errors such as omission/insertion of notes or wrong tone heights/durations. The recognition rate is maximized by applying evolutionary strategies for parameter optimizations. The parallel implementation on a Linux-based PC cluster and the computational effort are discussed. Quantitative results are presented for melody recognition. We compare the method with related work and conclude with an outlook