Short Description

Learning to speak again without a larynx is a challenging task. Usually speech produced by the bionic voice is more difficult to understand than using healty voice. One reason is, that the substituion voice is to quiet. Therefore, many use an amplifier to increase the assertiveness of their voice. Unfortunately, one problem many experience is the frequent occurence of feedback howling. This project should compare three different approaches to feedback reduction.
a) Feebback suppression: It feedback is detected the gain of the problematic frequency is reduced
b) Feedback reduction with frequency shift: A small frequency shift reduces the possibility of a loop gain > 1
c) Feedback cancellation: The transfer function between the loudspeaker and the microphone is estimated and the the feedback is subtracted.

Depending on the type of project either improvement of algrithms or only comparison can be performed. This project is perfectly suited for teamwork.

Your Tasks

• Combine the three algorithms into an easy to use combination with a portable loudspeaker.
• Design a user test to evaluate the different approaches.
• Compare the approaches concerning, robustness, reliability, computational complexity, …

Your Profile

• Motivation and interest in the topic
• Background in speech signal processing
• Background in either Matlab, Python or Julia

Contact

Martin Hagmüller (hagmueller@tugraz.at or 0316/873 4377)