Machine learning and the structure of bird song
Supervisor: Dr Dan Stowell
Research group(s): Centre for Digital MusicBirdsong is fascinating, not least because of its complex structure. What are the regularities, the common patterns? If we can understand the structure of birdsong we can help to decode the information being transmitted. Practically we can use such analysis for monitoring bird populations and automating animal behaviour studies. You will investigate how to adapt machine learning techniques to best represent the structure present in birdsong audio sequences. This could include deep learning (neural networks such as autoencoders, GANs, RNNs, WaveNet, embeddings), and/or probablistic models such as hidden Markov models. The immediate goal of such machine learning is to encode/generate bird song sequences and to produce representations that lead to good recognition performance. This topic is suitable for someone with a background in a mathematical topic such as machine learning or statistical optimisation, ideally with a passion for sound.