Geometric Analysis of Musical Audio Data

Project Summary

In this work, we turn musical audio time series data into shapes for various tasks in music matching and musical structure understanding. 

Themes and Categories
Year
Contact
Christopher Tralie
Electrical and Computer Engineering
chris.tralie@gmail.com

In particular, we use sliding window representations of chunks of audio to create high dimensional time-ordered point clouds, and we extract information by analyzing the geometry of these clouds.  We have shown, for example, that sequences of these shapes can be used to identify two different versions of the same song, or "cover songs."  We have also shown that both local and global musical properties can be expressed in geometric language.  For instance, hip hop is very "wiggly" while classical music is very "smooth."  Choruses and verses tend to live in distinct clusters connected by paths, and "bridges," or detours to a different musical idea, show up as large loops.

Related People

Related Projects

The visibility of hate groups such as the Alt-Right became mainstream into contemporary political culture during the Unite the Right Rally in Charlottesville, VA in 2017. This project aims to explore methods to quantify the presence of Latinxs within the Alt-Right, particularly in how they racialize themselves in a space that often spews hate towards Mexicans and other marginalized groups from Latin America. Using data from multiple sources (such as Twitter, Stormfront, and Breitbart), we developed a corpus of tweets, subthreads, and articles, and analyzed this data using basic natural language processing (NLP) techniques.

Project Lead: Cecilia Márquez

Project Manager: Susan Jacobs

 

Click here to view the team's project summary slides

 

Watch the team's final presentation (on Zoom) here:

We apply word embedding models to corpora from the start of the Early Modern period, when the market economy began to dramatically expand in England. Word embedding models use neural networks to map vectors to words so that semantic relationships are preserved within the vectors’ geometry. Such models have been successful in understanding cultural trends and stereotypes in large corpora of texts, but these techniques are infrequently used on texts dating much farther back than the 19th century. Using newly developed methods for analyzing word embeddings, we track the development of the meanings of words related to consumerism, including their relationships with gender over time.

 

Project Leads: Astrid Giugni, Jessica Hines

Project Manager: Chris Huebner

 

Click here to view the team's final poster

 

Watch the team's final presentation (on Zoom) here:

Led by Dr. Eva Wheeler, this project considers how racial language in African American literature and film is rendered for international audiences and traces the spread of these translations. To address the study’s primary questions, the team analyzed a preliminary dataset and explored the relationship between translation strategy and different categories of racial language. The team also conducted a macro-level analysis of the linguistic, temporal, and geographic spread of African American stories using the IMDB and WorldCat databases. We have found a large amount of variation in how African American stories are rendered, which can in part be explained through a social scientific lens.

 

Project Lead: Eva Wheeler

 

Project Manager: Bernard Coles

 

Click here to view the team's project poster

 

Watch the team's final presentation (in Zoom) here: