Adolfo Balderas’s senior project: Hardware for voice anonymization

Engineering Design Projects (ES 100), the capstone course at the Harvard John A. Paulson School of Engineering and Applied Sciences (SEAS), challenges seniors to engineer a creative solution to a real-world problem.

Real-Time Voice Encoder for Audio Privacy

Adolfo Balderas, S.B. ‘26, Electrical Engineering

Advisor: Jason Yik

• Please give a brief summary of your project.

My project was on voice anonymization. I was using an field-programmable gate array (FPGA), which is a piece of hardware used to encode various algorithms to do digital signal processing on an audio input, namely a person's voice. I was feeding voices into the FPGA and then outputting a processed signal that implemented pitch- and format-shifting. 

• What real-world challenge does your project address?

A lot of applications in today's day and age are using our voice. We have digital assistants like Siri and Alexa. We are on Zoom. Our voice carries a lot of personally identifiable, biometric data, and so privacy is very important. I was focusing on analyzing the trade-offs and finding some good metrics for balancing privacy and intelligibility. You want to be able to achieve a point that is simultaneously private but still able to be understood. There's no point in having garbled audio if no one understands it. The flip side is if you have a very clear audio signal that anyone can understand, but it's not secure and very easy to distinguish who is speaking and what they’re saying. 

• What did you conclude?

From all the testing that I did, I couldn't find an ideal combination of the different algorithms and trade-offs that is perfect for all the speakers that I was testing. My conclusion is that there's a lot of content and speaker dependency. What is being said matters just as much as who is saying it, and that influences both intelligibility and privacy.