The weekly SILO Seminar Series is made possible through the generous support of the 3M Company and its Advanced Technology Group


with additional support from the Analytics Group of the Northwestern Mutual Life Insurance Company

Northwestern Mutual

The Least-Squares Invertible Constant-Q Spectrogram (LSICQS) and Its Application to Phase Vocoding | Human Semi-Supervised Learning

Atul Ingle | Bryan Gibson, Graduate students in ECE and CS respectively

Date and Time: Mar 14, 2012 (12:30 PM)
Location: Orchard room (3280) at the Wisconsin Institute for Discovery Building


*** Atul's talk:
Title: The Least-Squares Invertible Constant-Q Spectrogram (LSICQS) and Its Application to Phase Vocoding

Building on the idea of a constant-Q transform popularized by Judith Brown in the 1990's, we will develop a constant-Q
spectrogram representation which is invertible in a least-squares sense. We will see that the LSICQS overcomes the
time decimation issue of a regular constant-Q spectrogram so that there is some hope of obtaining a good inverse.
Next, we will design a phase vocoder using the LSICQS by handling a few subtleties related to phase reassignment.
Finally, I will regale the audience with a couple of audio examples.

*** Bryan's talk:
Title: Human Semi-Supervised Learning

Current psychological models of human categorization are almost exclusively supervised, in that they only make use of labeled data while ignoring unlabeled data. This talk will discuss a set of experiments showing that human behavior, in a categorization task, is effected by both labeled and unlabeled data. Using equivalences between models found in human categorization and machine learning research, semi-supervised models developed in machine learning which make use of both labeled and unlabeled data can be applied to human learning. The resulting new models prove useful for explaining human behavior in these mixed, labeled and unlabeled, task settings.