Probabilistic Characterization and Synthesis of Complex Driven Systems

7/28/00


Click here to start


Table of Contents

Probabilistic Characterization and Synthesis of Complex Driven Systems

PPT Slide

outline

why bother building a digital violin?

…violin model as a benchmark test for algorithms

context and history of an idea: embedding synthesis

embedding synthesis (2) : why we need a more sophisticated representation?

available machine learning tools

… need for a better learning tool

Part I Cluster-weighted modeling

non-linear function approximation

cluster-weighted modeling (CWM)

model architecture (1)

model architecture (2): forecast of y and ?y

model estimation: expectation-maximization (EM) (Dempster et al. ’77)

EM for Gaussian mixtures (1)

EM for Gaussian mixtures (2)

evaluation and system characterization

evaluation and characterization (2)

a cluster-weighted input-output hidden Markov model (1)

a cluster-weighted input-output hidden Markov model (2)

PPT Slide

Part II Synthesis architectures and applications

classification theory

cluster-weighted classification

application I: stimulus detection from MEG data

application II: consumer fraud characterization

application III: electronic noses

prediction and estimation theory

application I: prediction of the M3-competition

application I: prediction of the M3-competition data (2)

application II:microwave device characterization

microwave device characterization (2)

Part III Data-driven modeling of musical instruments

why would we want digital instruments?

musical synthesis

the digital stradivarius

modeling sequence

sensor hardware - schematic

PPT Slide

interfaces

input patterns

analysis patterns

representations (1): sinusoidal analysis-synthesis

representations (2) - wavetable synthesis ? cluster-weighted sampling

cluster-weighted sampling (2)

experimental results : sinusoidal synthesis

experimental results : wavetable and mixed synthesis

contributions (1)

contributions (2)

conclusions and future work

acknowledgements (1)

acknowledgements (2)

Ende

Author: Bernd Schoner