Signal Processing Researchers Awarded the Best Paper Award
The article has become one of the cornerstone papers on automatic recognition of environmental sounds. The paper proposed a deep learning architecture based on convolutional and recurrent neural networks that simultaneously enable learning representations that allow recognizing a wide variety of different kinds of sound events, and modeling large temporal context which is needed to disambiguate various acoustically similar sound events. The generality of the approach has enabled applying it successfully to a wide range of tasks, ranging from bioacoustic monitoring and smart homes to smart cities. The paper has also set an example of open science and public benchmarks for the whole research community on environmental sound recognition.
The research resulting to the article was done at the ERC project “Computational Analysis of Everyday Soundscapes” led by professor Tuomas Virtanen in Tampere University of Technology. Virtanen and Heittola are currently working in the Audio Research Group of Tampere University, and Emre Çakır at Inscripta, Giambattista Parascandolo at OpenAI, and Heikki Huttunen at Visy. The award will be given at the ICASSP conference in Seoul in April. The IEEE’s first society, the Signal Processing Society is the world’s premier professional society for signal processing scientists and professionals since 1948, and has more than 20 000 members.