We use cookies to help us to improve our site and enables us to deliver the best possible service and customer experience. By clicking accept or continuing to use this site you are agreeing to let us share your data with third parties in accordance with our privacy policy. Learn more

News

From theory to practice: Dave Betts describes how to build a successful commercial audio research team

Jan 3, 2022

Good teams get their strength from diversity, and a commercial research team is no exception to that rule. So, when we built the research team here, we weren't interested in hiring lots of ‘cookie cutter’ researchers – we wanted to have a team which included a wide breadth of experience within it. We knew that the best way to enlarge our capability as a team was to hire people from different industries and with different research backgrounds. 

What does this mean in practice? The team started with just two of us – Mohammad and I, who worked together for ten years on the development of our blind source separation algorithms. Even with just two of us, the team was diverse: Mohammad has a very strong understanding of the base of the principles behind blind source separation, with a PhD on blind source separation using beamforming techniques. On the other hand, I’m fundamentally a Bayesian mathematician; as such, I’m very keen on mathematical theory and on there being some rigour behind the processes that are involved. But I’m not the only mathematician on the team: James has an Applied Maths PhD in nonlinear and musical acoustics from Cambridge. 


Yesenia’s experience since her PhD has included time with mobile phone companies, using binaural techniques and psychoacoustics to get optimum sound reproduction from tiny mobile phone speakers. So she brings us knowledge of audio reproduction and processes on phones, which is invaluable. Tom has worked in machine learning and artificial intelligence as well as audio, bringing to the table experience in mobile voice communications (echo cancellation, multi-microphone noise reduction) as well as smart home applications for far-field voice applications. Sergi has a good background in acoustic scene classification and sound event classification with neural networks. We are of course all familiar with each other’s areas, but because overall the team has a number of different centres of expertise, together our knowledge covers the widest possible area of audio knowledge and its application to real world problems. 


How do we work? Everyone in the team wants to increase their knowledge and deepen their experience so as a team we’re always growing, always absorbing new knowledge, always taking on new problems. We don't pigeonhole researchers: to help them develop their careers we will always, where possible, give people projects and jobs that are actually slightly outside their comfort zone. The objective is for them to not just complete the project, but to learn from others in the team the appropriate techniques for that particular field. And by taking in other people's ideas and then merging it with their own creates a real cross-fertilization of ideas.


There are three important considerations which we take into account when evaluating our research and our proposed solutions. The first is that it’s not enough simply to have something that works - we must understand why it works. Some commercial researchers think, “ If it works, it works. You don't need to understand why it works,” but if you take that attitude, then what happens when the next problem comes along? We try to get a really good understanding of why something works because that's how you move on, that's how you make improvements. 


That fundamental understanding allows you to create the next variation which then will solve a new problem. And it means that we'll be able to solve other problems in the future well, keeping our research where it needs to be: at the cutting edge. 


However, just having the maths and the research rigour is not enough. Our second consideration is that we must always think about our solutions from the point of view of the customer. As a commercial audio researcher, you do need to be able to bridge the gap between the maths and the ideas behind the processes, and what's needed in the commercial real world. If our technology doesn’t solve a pain point for our customers, they won’t be interested in incorporating it into their products. And if it isn’t practical for them to do so, they won’t buy it. So - always bear in mind what your end user wants, what they need and the overall practicalities of how your solution is going to be used. 


There's a lot of nice maths behind our algorithms but having nice maths doesn't matter if you don't get a nice sounding result. Because we’re working in audio our third important consideration is the end user listening experience – clear, natural sounding audio with no horrible artefacts or distortions. It doesn't matter how brilliant your research is if it gives a result that doesn't work the way that the user wants it to work. So we are always keeping the consumer at the forefront of everything we do, and testing our proposed solutions against the desired user experience. 


To summarise – to build a successful commercial research team you should firstly ensure you have a team with different individual specialisms, but preferably including commercial experience. Give them interesting assignments that keep them learning and growing, and to keep at the forefront of your research area, make sure you understand why things work, to allow you to develop the next iterations of your products. Finally, keep the team constantly focussed on what your customer and your customer’s end users actually need.