CCRMA
next up previous contents
Next: Audification of Data Up: Research Activities Previous: Audio Signal Processing

Machine Recognition in Music


Subsections

Joint Estimation of Glottal Source and Vocal Tract for Vocal Synthesis Using Kalman Smoothing and EM Algorithm

Pamornpol (Tak) Jinachitra

In this research, a joint parameter estimation of the derivative glottal source waveform and the vocal tract filter is presented in which aspiration noise and observation noise are taken into account within a state-space model. The Rosenberg-Klatt glottal model is used in conjunction with an all-pole filter to model voice production. The EM algorithm is employed to iteratively estimate the model parameters in a maximum-likelihood sense, utilizing a Kalman smoother in the expectation step. The model and estimator allow for improved estimates of model parameters for re-synthesis, yielding an output which sounds natural and flexible for modification.

Bayesian Modeling of Musical Expectations via Maximum Entropy Stochastic Grammars

Randal Leistikow, Harvey Thornburg, and Jonathan Berger

When presented with musical sounds, humans take advantage of prior knowledge of acoustic and musical context to accomplish an impressive array of music cognition tasks, such as meter tracking, transcription, style classification, instrument identification, harmonic analysis, and melodic prediction. This project focuses on the development of an extensible dynamic Bayesian inference engine capable of performing such tasks on symbolic data. In order to model listeners with differing prior expectations, we have developed a method of automatically transforming music-theoretic rule sets into maximum entropy stochastic grammars suitable for use in dynamic Bayesian networks for modeling signal-level data, such as the joint onset detection, transient region identification, and melody transcription framework of Thornburg, Leistikow and Berger. Encoding rule-based expectations allows the system to infer which rules are active at each point in a piece of music and to identify which rules are violated at points of musical surprise.

Automatic Transcription of Polyphonic Piano Music

Randal J. Leistikow

The goal of this project is to develop a system that accepts as input a single-channel or stereo recording of solo piano music and outputs a data file containing the performance parameters necessary to resynthesize an expressively realistic performance of the music. Applications of such a system include the abilities to ``resurrect'' performances from historic recordings and hear them played live on a modern reproducing piano, release pristine versions of recordings corrupted by noise, process the output data to study aspects of performance practice or playing style, and code piano music extremely efficiently.



© Copyright 2005 CCRMA, Stanford University. All rights reserved.