1. A music retrieval system using chroma and pitch features based on conditional random fields
- Author
-
Takuya Fujishima, Miki Arai, Kouhei Sumi, and Hashimoto Seiichi
- Subjects
Conditional random field ,business.industry ,Computer science ,Speech recognition ,Feature extraction ,Pattern recognition ,Data modeling ,ComputingMethodologies_PATTERNRECOGNITION ,Robustness (computer science) ,Music information retrieval ,Artificial intelligence ,business ,Hidden Markov model ,CRFS ,Independence (probability theory) - Abstract
This paper presents a new symbol-based retrieval method on a polyphonic music collection which takes a sequence data of users' performances as a query. We focus on chroma and pitch features to yield a robust retrieval with queries which are generated from different arrangements and which include some mistakes. Conditional random fields (CRFs) are used to enhance simultaneous utilization of chroma and pitch features. This is because CRFs can discriminate the correct sequence from all the other candidate sequences without independence assumptions for features of the inputs. Experimental results show that the use of multiple features based on CRFs leads to a significant improvement of retrieval accuracy and accomplishes robust music retrieval regardless of performance style of queries.
- Published
- 2012