Back to projectsVolver a los proyectos

My Thesis!

Voice Quality Modification using the WORLD Vocoder

Project imageImagen del proyecto

In 2021 I finished my thesis and could finally graduate as an Electronics Engineer. In this little article I want to tell you grosso modo the motivation behind the work I did, the challenges I faced, and the final result.

If you're interested, you can read the full thesis or the associated paper. Unfortunately, I had a massive case of burnout when I finished the paper, so I never made the corrections and it never got published 😥.

Motivation #

When you, dear reader, or any healthy person speaks, your body does something surprising. It's constantly monitoring how you speak, and it makes small adjustments to your voice so it matches what you expect your voice to sound like. Maybe you're speaking at a lower pitch, or speaking slightly louder than you meant to. So, as part of this control mechanism of your body, you'll unconsciously correct that. You'll raise your pitch, speak more slowly, etc.

For example, the Lombard effect has been described in the literature, where speakers involuntarily tend to raise their voice when they speak in a noisy place. This effect doesn't only happen with pitch, but also happens with other characteristics of the voice.

Now, here comes the interesting part. There are certain patients who aren't able to correctly correct their voice based on the feedback. For example, people with Parkinson's, in test settings where they're given artificial, real-time feedback of their voice, incorrectly regulate the frequency of their voice (citation needed, honestly I'm not going to look it up, this is an informal article, dear reader).

The goal of this thesis was to create an algorithm that could modify a speaker's voice quality (we're no longer talking about pitch or frequency, but something much more complex), in real time, with the future aim of being able to measure how people respond to that internal monitoring when their voice quality is modified.

Challenges of the thesis #

Well, you might say, the problem sounds simple, why should I care about this? The reason is that the explanation above leaves a crucial, super important concept hanging. What is voice quality? And the thing is, we could say that's what I spent two years asking myself.

(WIP, forgive me, dear reader, I haven't finished writing this article yet)

Under Construction