Labeling a user’s speech in real-time for always-on VoIP

TurnMeUp is an iOS app for always-on voice communications. Users leave the app running in the background and can talk to the recipient (also using the app) at any time. This app is especially useful for coworkers listening to their own music in the background without needing to enter and exit voice call sessions manually. To conserve bandwidth and ensure that users listen to music without being unnecessarily interrupted, TurnMeUp sends voice signals to the recipient only if the user is speaking. The purpose of this project is to improve the algorithm used for detecting when the user is speaking (as opposed to background noise or other people speaking) in real time, using contemporary machine learning methods. This project will potentially improve the performance of TurnMeUp; to the greatest extent possible, it will ensure that the entirety of a user’s speech, and nothing else, is sent to the recipient.

Willie Chang
Faculty Supervisor: 
Frank Rudzicz
Partner University: