PhD studentships in AI for Sound

The AI for Sound project in the Centre for Vision, Speech and Signal Processing (CVSSP) at the University of Surrey is offering the following PhD studentships in AI for Sound, available from 1 October 2021:

• Automatic sound labelling for broadcast audio: The aim of this project is to develop new methods for automatic labelling of sound environments and events in broadcast audio, assisting production staff to find and search through content, and helping the general public access archive content. The project will undertake a combination of interviews and user profiling, analysis of audio search datasets, and categorisation by audio experts to determine the most useful terminology for production staff and the general public as user groups. The project will develop a taxonomy of labels, and examine the similarities and differences between each group. The project will also investigate the application of a labelled library in a production environment, examining workflows with common broadcast tools, then integrating and evaluating prototype systems. The project will also investigate methods for automatic subtitling of non-speech sounds, such as end-to-end encoder-decoder models with alignment, to directly map the acoustic signal to text sequences. Working with BBC R&D, the student will develop software tools to demonstrate the results, especially for broadcasting and the management of audiovisual archive data, and benchmark the results against human-assigned tags and descriptions of audio content. Using archive data provided by BBC R&D, the student will engage with audio production and research experts through Expert Panels, and potential end users through Focus Groups. As part of this PhD, you will have the opportunity for close day-to-day collaboration with the BBC as a member of the R&D Audio Team.
• Information theoretic learning for sound analysis (Funding Eligibility: UK applicants only): The aim of this PhD project is to investigate information theoretic methods for analysis of sounds. The Information Bottleneck (IB) method has emerged as an interesting approach to investigate learning in deep learning networks and autoencoders. This project will investigate information-theoretic approaches to analyse sound sequences, both for supervised learning methods such convolutive and recurrent networks, and unsupervised methods such as variational autoencoders. The project will also investigate direct information loss estimators, and new information-theoretic processing structures for sound processing, for example involving both feed-forward and feedback connections inspired by transfer information in biological neural networks.

Application Deadline: 1 August 2021

CVSSP also has a number of ongoing PhD studentship opportunities for outstanding PhD candidates in all aspects of audio-visual signal processing, computer vision, and machine learning, including for research related to machine learning and audio signal processing. We also welcome enquiries from self-funded and part-funded candidates.