| Abstract: |
Computers are playing an increasingly important role in creating, analyzing, and accessing music. Intelligent interaction depends on the computer ``understanding'' the musical signal, which concretely means automatically extracting information such as onset locations, pitch, timbre, musical structure, and genre. This is known generically as music information retrieval (MIR).
Up to now, MIR research has rarely addressed non-Western music. In addition to expanding the scope of MIR tools, cross-cultural research has important implications for general theories of music and music cognition. This talk will discuss two important problems in MIR for Indian classical music (ICM): automatic transcription of tabla music and raag identification.
Tabla is a ubiquitous percussion instrument in ICM capable of producing a wide variety of timbres and rhythms. Automatic transcription will allow novel analyses to be performed as well new music to be created based on tabla patterns. The automatic transcription system recognizes different tabla strokes using a statistical pattern recognition framework. Recorded performances are automatically segmented into individual strokes by looking for discontinuities in phase and amplitude. Spectral and temporal features are then calculated on those strokes, which are used to train multivariate, neural network, and tree-based classifiers. Approximately 15 stroke categories could be reliably distinguished. Rhythm analysis is performed using autocorrelation and duration histogram-based approaches.
Raag is the central melodic concept of ICM. Raags are complex melodic structures composed of basic phrases and methods for elaborating them. Here, raag classification is attempted using pitch-class distributions and pitch-class dyad distributions calculated on short (30 sec.) segments of performances. First segments are pitch-tracked using a variant of the Harmonic Product Spectrum algorithm, after which they are automatically segmented into notes. Pitch-class labels are assigned to each note from which the distributions for each segment are computed. Again, using a statistical pattern recognition framework, the system was successful in identifying 17 different target raags with a high degree of accuracy (> 90%).
|
| Bio: |
Parag Chordia is an assistant professor of music in the College of Architecture. He is part of the music technology group, and specializes in Music Information Retrieval (MIR) research and applications. Through his research, Dr. Chordia attempts to synthesize advances in pattern recognition and signal processing to create systems that can 'listen' intelligently. He is particularly interested in creating tools that can be used to advance research in computational music theory and music cognition. Dr. Chordia is also interested in the application of MIR tools for composition and multimedia performance. His own compositional work draws on both Indian classical and electronic music traditions.
Dr. Chordia received his PhD in media 'Computer-based Music theory and Acoustics' from Stanford University's CCRMA, and his BA in Applied Mathematics from Yale University.
Before turning to academia, he founded Bol Records, an Indian classical music label, where he served as CEO and artistic director. Most recently, he co-founded Yaari.com, a social networking tool for South Asians, where he served as CTO. Additionally, he is an active performer of Indian classical music and a disciple of the legendary Pandit Buddhadev Das Gupta.
|