Consultation des notices

In this paper we introduce a simple and fast method for realtimerecognition of multiple-pitches produced by multiple musicalinstruments. Our proposed method is based on two important facts:one that timbral information of any instrument is pitch-dependantand two, that the modulation spectrum of the same pitch seems toresult into a persistent representation of the characteristics ofthe instrumental family, as discussed in the paper. Using thesebasic facts, we construct a learning algorithm to obtain pitchtemplates of all possible notes on various instruments and thendevise an online algorithm to decompose a realtime audio bufferusing the learned templates. The learning and decomposition proposedhere are inspired by non-negative matrix factorization methods butdiffer by introduction of an explicit sparsity control. Our testresults show significant recognition rate for a realtime system andon real music recordings. We discuss further improvements that canbe made over the proposed system.