Robust Automatic Speaker Diarisation of Audio Documents by Exploiting Prior Sources of Information. Speaker Diarisation, the task of determining who spoke when, is a technology fundamental in deriving intelligent information from audio and multimedia resources. The requirement for efficient and accurate Speaker Diarisation systems, portable across different domains is heightened by the explosive growth of audio and multimedia archives online and throughout the world. This research will provide t ....Robust Automatic Speaker Diarisation of Audio Documents by Exploiting Prior Sources of Information. Speaker Diarisation, the task of determining who spoke when, is a technology fundamental in deriving intelligent information from audio and multimedia resources. The requirement for efficient and accurate Speaker Diarisation systems, portable across different domains is heightened by the explosive growth of audio and multimedia archives online and throughout the world. This research will provide the foundation for a commercial service of automatic Speaker Diarisation to be developed, growing Australia's impact on the information and communications technology (ICT) sector. The outcome of this research will also assist in the tracking of terrorist and unlawful activity by enabling effective intelligence gathering from different audio sources.Read moreRead less
Audio Visual Speech Recognition. Even though significant advances have been made in automatic speech recognition using acoustic information, the recognition accuracies are still poor in noisy and hostile environments such as in crowds, traffic, factory floors etc. In many of these applications visual information is or can easily be made available in addition to the audio. The aim of this project is to achieve an order of magnitude improvement in speech recognition accuracies in adverse environme ....Audio Visual Speech Recognition. Even though significant advances have been made in automatic speech recognition using acoustic information, the recognition accuracies are still poor in noisy and hostile environments such as in crowds, traffic, factory floors etc. In many of these applications visual information is or can easily be made available in addition to the audio. The aim of this project is to achieve an order of magnitude improvement in speech recognition accuracies in adverse environments by joint processing and modelling of the acoustic modality with visual information in the form of lip shapes and movements. The outcomes will be useful in human computer interaction in adverse environments as well as in the transcription and mining of multimedia data.
Read moreRead less
Dynamic Scheduling and Stochastic Control for Sensor Networks. Sensor networks are rapidly becoming important in applications from environmental monitoring, navigation to border surveillance. However, due to bandwidth constraints, even very simple networks have proven to be very complex to properly control. It is now necessary to efficiently allocate the 'limited available bandwidth' to sensors in order to share the most valuable data over the network. Therefore, this project proposes new techn ....Dynamic Scheduling and Stochastic Control for Sensor Networks. Sensor networks are rapidly becoming important in applications from environmental monitoring, navigation to border surveillance. However, due to bandwidth constraints, even very simple networks have proven to be very complex to properly control. It is now necessary to efficiently allocate the 'limited available bandwidth' to sensors in order to share the most valuable data over the network. Therefore, this project proposes new techniques using concepts of dynamic sensor scheduling and stochastic control to provide computationally feasible and near optimal solutions to the limited and varying bandwidth problem. This work will greatly enhance the operational performance of distributed sensor networks.Read moreRead less
The next generation speaker recognition system. The next generation of speaker recognition technologies developed through this project will enable secure person authentication by voice in financial transactions and benefit the community through the elimination of identity fraud. This project will safeguard Australia by identifying criminal suspects using their voice and combat terrorism by using voice to locate and track terrorists.
Robust, valid and interpretable deep learning for quantitative imaging. One of the biggest challenges in employing artificial intelligence is the “black-box” nature of the models used. This project aims to improve the effectiveness and trustworthiness of deep learning within quantitative magnetic resonance imaging. Deep learning has great promise in speeding-up complex image processing tasks, but currently suffers from variable data inputs, predictions are not guaranteed to be plausible and it i ....Robust, valid and interpretable deep learning for quantitative imaging. One of the biggest challenges in employing artificial intelligence is the “black-box” nature of the models used. This project aims to improve the effectiveness and trustworthiness of deep learning within quantitative magnetic resonance imaging. Deep learning has great promise in speeding-up complex image processing tasks, but currently suffers from variable data inputs, predictions are not guaranteed to be plausible and it is not clear to the end user how reliable the results are. The outcomes intend to deliver advanced knowledge and capability in artificial intelligence and machine learning that Australia urgently needs to capitalise on bringing deep learning into practical applications delivering economic, commercial and social impact.Read moreRead less
Automatic audio segmentation, classification, identification, search and retrieval. The research aims to develop generic tools for automated audio segmentation, classification, identification and search, with lowest possible computational complexity and highest accuracy and speed. The tools will be applicable to audio archive management, search of audio material over WWW and personal archives of music and audio-assisted video analysis. The industry will use the tools for automated broadcast ve ....Automatic audio segmentation, classification, identification, search and retrieval. The research aims to develop generic tools for automated audio segmentation, classification, identification and search, with lowest possible computational complexity and highest accuracy and speed. The tools will be applicable to audio archive management, search of audio material over WWW and personal archives of music and audio-assisted video analysis. The industry will use the tools for automated broadcast verification and identification for copyright surveillance and calculation of royalty payments, aiming to penetrate both Australian and overseas markets. The area of real-time audio scene analysis is in its infancy and the research aims to make significant contributions to this area.Read moreRead less
Multi-Modal Dictionary Learning for Smart City Operation and Management. This Project aims to provide new digital asset management tools for city councils to improve city services by utilising new sensing and automated learning technologies for recognising, tracking and auditing of assets. Currently, there are no digital tools available to handle these services. This project proposes new multi-modal sensing and mapping of city asset techniques by building new multi-modal dictionary learning proc ....Multi-Modal Dictionary Learning for Smart City Operation and Management. This Project aims to provide new digital asset management tools for city councils to improve city services by utilising new sensing and automated learning technologies for recognising, tracking and auditing of assets. Currently, there are no digital tools available to handle these services. This project proposes new multi-modal sensing and mapping of city asset techniques by building new multi-modal dictionary learning procedures. The new framework will recognise different conditions of city assets in real-time to make decisions. Expected outcomes of this Project include integration and easy access of assets with unique digital identities to help city councils, governments, and navigation services for real-time asset monitoring.Read moreRead less
Development of fundamental perception technology and algorithms for mining safety. The project will push the boundaries of mining safety research to deliver innovative and powerful tools to understand and control the level of risk of an operation. This knowledge will be used to develop algorithms to best assess safety issues in different scenarios, to design safety procedures and to develop operator training.