Algorithms may improve user-generated recordings
21 October 2015
Acoustic scientists at the University of Salford have developed new algorithms to improve user-generated recordings that are often poor, distorted or noisy.
A team led by Professor Trevor Cox asked thousands of volunteers to explain what they thought was interfering with the quality of sound on clips recorded in living rooms, on the street and at gigs, including at the Glastonbury Festival.
“People are often disappointed when they play their recordings back, after a concert or a party, but there is a real lack of understanding as to why,” explains Cox, professor of acoustic engineering and author of Sonic Wonderland.
“It could be microphone handling noise, distortion, wind noise or a range of other conditions. What we have worked out is a way of automatically assessing the relative impact of these sound errors.”
The algorithms, which makes it possible to tag content and quality, has already been applied to an app for assessing wind noise, which alerts the user when there is significant risk of the sound being affected.
The three-year Good Recording Project, led by Salford University, is a response to increasing demand from consumers and from broadcasters who often use amateur footage which is compromised by sound quality.
“We’re used to having visual processing improving our photos, such as the camera that spots faces and changes exposure, but we have not had the same tools to do the audio equivalent," adds Cox.
Rapid quality assessment could determine whether the sound is of broadcast quality without time consuming manual auditioning.
The £0.5 million project was funded by the Engineering and Physical Sciences Research Council and run in collaboration with BBC R&D and The British Sound Archive. The research is published in the journal, PloS One and in the Audio Engineering Society Journal and will be presented at the Audio Engineering Society Conference in New York (October 29 – Nov 1).