Randal Leistikow, Ph.D.


Center for Computer Research in Music and Acoustics
Stanford University

Research Dissertation Other Publications

Current Projects:

I am currently managing AV development and MIR research at Smule, a company working to bring the world together through music.

Research Interests:


"Bayesian Modeling of Musical Expectations via Maximum Entropy Stochastic Grammars"
When presented with musical sounds, humans take advantage of prior knowledge of acoustic and musical context to accomplish an impressive array of cognitive listening tasks, such as meter tracking, transcription, style classification, instrument identification, harmonic analysis, and melody prediction. This dissertation presents a dynamic Bayesian framework for modeling listeners with differing musical expectations.

Although a simulated listener with specific experience may simply be created by learning prior distributions directly from a given musical corpus, a more interesting approach is to construct a listener whose expectations are governed by rules of music theory. Such rules are often expressed as statements involving musical tendencies, e.g., "A large upward melodic interval is typically followed by a smaller downward interval." This dissertation focuses on a novel method of transforming music-theoretic rule sets into parameterized, maximum entropy rate distributions suitable for use in dynamic Bayesian networks. Encoding rule-based expectations allows the system to infer which rules are most responsible for predicting musical attributes at each time in a piece, and to identify which rules are violated at points of musical surprise.

In addition to enabling a wide variety of interesting musical tasks to be performed using symbolic data as input, our framework can also be integrated into compatible probabilistic models that use recorded audio signals as input. The signal processing layers encode acoustic expectations by modeling the spectrotemporal evolution of instrument tones, and segment the signal into a sequence of note events. A system in which signal and symbolic layers inform one another is desirable because musical expectations can help the system compensate for corrupted signals, and the ability to predict musical sequences suggests a future sequential Monte Carlo inference implementation in which sampling distributions concentrate on the most likely transitions, thereby avoiding the computational cost of evaluating all points in the potentially vast space of possible transitions.

Download: PDF

Other Publications:

Research Dissertation Other Publications