Two nonnegative matrix factorization methods for polyphonic pitch transcription

Emmanuel Vincent (1) , Nancy Bertin (2) , Roland Badeau (2)

1 METISS - Speech and sound data modeling and processing
2 Télécom ParisTech

Polyphonic pitch transcription consists of estimating the onset time, duration and pitch of each note within a music signal. Adaptive signal models such as Nonnegative Matrix Factorization (NMF) appear well suited to this task, since they can provide a meaningful representation whatever instruments are playing. In this paper, we propose a simple transcription method using minimum residual loudness NMF, harmonic comb-based pitch identification and threshold-based onset/offset detection, and investigate a second method incorporating harmonicity constraints in the NMF model. Both methods are evaluated in the framework of MIREX 2007.

Domaines: Traitement du signal et de l'image [eess.SP]; Traitement du signal et de l'image [eess.SP]

Emmanuel Vincent, Nancy Bertin, Roland Badeau. Two nonnegative matrix factorization methods for polyphonic pitch transcription. 2007. ⟨inria-00544213⟩