Computational modeling of phonetic learning and adaptation using probabilistic inference approaches


Connor Mayer
Xin Xie

Lecture times

MTWRF 10:30 am-12:00 pm

Lecture locations

T SSPB 1222

Course Description

This course will cover computational models that learn linguistic structure from phonetic data using probabilistic inference techniques. In part 1 of the course, taught by Connor Mayer, we will focus on the task of learning vowel categories from acoustic input using Gaussian mixture models and Gibbs sampling. In part 2 of the course, taught by Xin Xie, we will look at models of structured variability in speech, focusing on stop categories. In addition to the technical details of these models, we will discuss how they can serve as useful complementary tools to experimental work.

Course format

Part 1 of the course will be taught in Python using Google Colab.

Part of 2 the course will be taught using R. We recommend that you install RStudio and R Markdown.

Intended audience

We assume basic familiarity with Python, R, or another programming language.


8/1 A crash course in phonetics Code (blank) Code (solutions)
8/2 Modeling vowel acoustics using Gaussian mixture models Code (blank) Code (solutions)
8/3 Learning vowel categories using Gibbs sampling Code (blank) Code (solutions)
8/4 Modeling speech perception as probabilistic inference
8/5 Modeling adaptive changes in perception