Meet the researcher – Mark Plumbley
In the first of a regular series of Q&As with our researchers, the University of Surrey’s Mark Plumbley talks about his hopes for the Multi Modal Models working group and his love of sound and music.
Mark Plumbley,
University of Surrey
“Work in AI is so broad and fast-moving, the AI Hubs help people to keep up with what is going on.”
Tell me about yourself and your work
I work in AI for sound - so that is using machine learning, artificial intelligence models to analyse sound, generate sound, separate different sounds. It is quite an interesting area, because most other AI researchers are working on things like languages or images, so it's a nice little corner of the space that I'm working in.
We have a small group working on sound at the University of Surrey as part of the Centre for Vision, Speech and Signal Processing (CVSSP), and also part of our pan-University Institute for People Centred AI (PAI). What this means is that CVSSP works on vision and signal processing handling the technical side of learning, while PAI looks across broad applications and it can used as part of sound recording by arts and humanities and by creative people too.
Within my own group, I have two or three researchers as part of a fellowship and I have three or four PhD students that work with me on different aspects of sound and machine learning and AI. There are other collaborators within the group that we work with.
When you work with the hub, which working group are you part of?
I'm working with the working group on Multi Modal Models, although we're still at early stages of that. I'm bringing the audio aspect to the multi modal models. At Surrey we've been looking at text and audio multi-modal models. We’ve been asking: can you take text prompts and turn them into audio generated models? Or can you take audio and turn it into text descriptions?
But we have also recently been looking at things like audio visual generation from text. You can't just do the audio and visual generation separately. You need to generate the vision and the audio at the same time. Otherwise, for example, an image of a drumstick hitting a surface. You need the sound and the picture to match in terms of how they work.
What do you think the AI Hubs bring to the research ecosystem?
Work in AI is so broad and fast-moving, the AI Hubs help people to keep up with what is going on in everything that might be related to what you are interested in. AI Hubs, like the Gen AI Hub, bring together key researchers from institutions across the UK, in areas like multimodal generative AI, so we can collaborate on more challenging issues, instead of just working in silos, in individual groups.
So how did you end up as an AI researcher?
Well, I first got the beginnings of AI when I was an undergraduate. There was a third-year course with a researcher from Bell Systems technical laboratory that was talking about pattern recognition and AI. That really got me interested in the area, but the maths was a bit hard, so I had to ask a friendly mathematician to help along that side! After I had been in industry for a couple of years, I began to get an interest in speech recognition, and that began my first interest in doing a PhD in that area. And then the PhD became what was in the late 1980s the work on neural networks and connectionist models. So that's where I started, and I've moved through music and sound since then.
What would it surprise people to know about you?
I used to do quite a lot of choral singing, and so I've had the opportunity to be in the backing choir, the London Symphony Chorus in a couple of Proms concerts, and also in the recording of Paul McCartney Standing Stone. I’ve had some great opportunities to do, do some big singing. That is something I am trying to get back into at the moment.
