The Interactive Audio Lab

Welcome to the Interactive Audio Lab, headed by Bryan Pardo. We are in the Electrical Engineering and Computer Science Department of Northwestern University . We apply machine learning, signal processing, natural language processing, and database search techniques to make new auditory tools and interfaces.

Current Projects

A music search engine you can sing to
Audio Imputation
Estimate missing information in audio
Spatial Source Separation
Amplify sounds coming from a particular direction
Separation by repetition, by repetition, by repetition, ...
Adaptive User Interfaces
Don't learn the tool, let the tool learn you
Separate music sources in real time with the help of a musical score
Score Alignment
Learn about the music
Music Story
Make videos from music
Multi-pitch Estimation & Streaming
Track more than one pitch at once

Demos and Products

  • Reverbalize is a tool for manipulating reverberation using natural language. Watch a video here. Try it out here.
  • SynthAssist is a tool for programming audio synthesizers using using similar example recordings, such as vocal imitations. Watch a video here.
  • Mixploration is an audio mixing tool for finding diverse mixes. Watch a video here.
  • Just sing to Tunebot to find your song.
  • SocialEQ is a tool that learns the meaning of sound adjectives from you.
  • Toneboosters released TB EZQ, a zero-latency, VST-plugin implementation of our 2DEQ audio equilizer.

Data Sets

  • SocialReverb, a dataset which has definitions for words in terms of how to manipulate a reverberation tool. By using this dataset, you can make new sorts of interactions with parametric reverberators possible, such as "make it boomy" or "make it sound like it is in a church"
  • SocialEQ, a dataset which has definitions for words in terms of how to manipulate an equalization tool. Gives the relationship between a word like "warm" and how to change the equalization to make some audio "warm".
  • Bach10, a versatile polyphonic music dataset for Multi-pitch Estimation and Tracking, Audio-score Alignment and Source Separation.
  • Ground-truth Pitches for the PTDB-TUG speech pitch tracking dataset. The ground-truth pitches were calculated using Praat on the microphone signals.
  • Jazz Score/Performance Database, a database of MIDI recordings performed by professional Chicago jazz pianists using lead sheets, useful as training and validation data for a jazz score following program.