Symbolic pitch modeling is a way of incorporating knowledge about relations between pitches into the process of an- alyzing musical information or signals. In this paper, we propose a family of probabilistic symbolic polyphonic pitch models, which account for both the “horizontal” and the “vertical” pitch struc- ture. These models are formulated as linear or log-linear interpo- lations of up to fi ve sub-models, each of which is responsible for modeling a different type of relation. The ability of the models to predict symbolic pitch data is evaluated in terms of their cross-en- tropy, and of a newly proposed “contextual cross-entropy” mea- sure. Their performance is then m easuredonsynthesizedpoly- phonic audio signals in terms of the accuracy of multiple pitch estimation in combination with a Nonnegative Matrix Factoriza- tion-based acoustic model. In both experiments, the log-linear com- bination of at least one “vertical” (e.g., harmony) and one “hori- zontal” (e.g., note duration) sub-model outperformed a pitch-de- pendent Bernoulli prior by more than 60% in relative cross-en- tropy and 3% in absolute multiple pitch estimation accuracy. This work provides a proof of concept of the usefulness of model inter- polation, which may be used for improved symbolic modeling of other aspects of music in the future.
Authors
- dr inż. Stanisław Raczyński link open in new tab ,
- Emmanuel Vincent,
- Shigeki Sagayama
Additional information
- DOI
- Digital Object Identifier link open in new tab 10.1109/tasl.2013.2258012
- Category
- Publikacja w czasopiśmie
- Type
- artykuł w czasopiśmie wyróżnionym w JCR
- Language
- angielski
- Publication year
- 2013
