Description: Computer based sound production and music composition, synthesis of musical sound or speech by computer date back to the first days of
digital computation. However, despite recent technological advances in synthesis, compression, processing and distribution of digital audio, the simulation of a general purpose intelligence, that will enable machines to display human level listening abilities is still far from reality.
Machine listening deals with developing computational methods for simulating perceptual mechanisms underlying human hearing. As these mechanisms are only partially understood, this is a research topic that poses many scientific questions and computational challenges. Modern approaches to this problem require computational modelling of sound in conjunction with human listening. A long term engineering goal is to construct software that can listen, imitate and autonomously interact with humans.
This course will deal exclusively with analysis of musical audio and environmental sound, source separation and music modelling. Audio and music signals have a complex and nonstationary temporal structure, both on physical-signal and cognitive-symbolic level. The main challenge in listening is the identification and classification of individual sound events (e.g. notes or alarm sounds), their invariant characteristics (e.g. timbre) as well as extraction of higher structure and semantic information (e.g. tempo, harmony, expression) from a single time series.
This is truly a cross discipline subject as it requires information from several fields: signal processing (spectrum analysis, filtering, and audio transforms); psychoacoustics (sound perception); cognitive sciences (neuroscience and artificial intelligence); machine learning (statistics, feature extraction) acoustics (physics of sound production); and music (harmony, rhythm, timbre, form).
* Understanding computational aspects of problems in hearing as related to audio and music understanding
* Investigation of computational methodologies to solve audio and music analysis, classification, separation and synthesis problems
* Provide an overview of the current state-of-the-art in audio, music and acoustic processing applications such as transcription, music information retrieval, audio restoration, coding or source separation
The treatment of methodological issues in the course will be slanted towards pattern recognition, statistical machine learning and probabilistic modelling.