AI-driven Sound Analysis

The project will look into proposing new machine learning methods for sound analysis and synthesis. This includes separating different sound sources from mixed recordings, and remastering and synthesising new scenarios.

In line with the School of Computing guidance for PhD students, the student will mainly focus on literature review in the first four months and produce the first formal report at the end of the fourth month. The report should be a comprehensive literature review on the proposed research topic and also identify a few possible research directions.

At the end of the ninth month, the student will produce the first-year report which serves the purpose of the transfer viva by the end of the first year. The report will include the formal introduction and statement of the PhD topic, a comprehensive literature review, current work progresses and future plans for the remaining of the PhD.

The PhD progress will be reviewed annually from the second year.

The first stage of the research will focus on sound analysis, especially in how to isolate different sound sources from mastered/mixed audio files. Extracting different sounds from mixed audio files will be the main goal of this stage. New machine learning methods will be proposed for this purpose.

The second stage will focus on extending the sound extraction methods into different applications. Such applications might include re-mastering the sound tracks from existing recordings, synthesising new tracks with extracted sounds, dynamic sound generation for different applications, e.g. Virtual Reality.

The research fits squarely into EPSRC's research area 'Vision, hearing and other senses', spanning across multiple areas including Artificial Intelligence Technologies and Human-Computer Interaction, and Music and Acoustic Technology.