Recurrent Lateral Spiking Networks for Speech Enhancement
Automatic speech recognition accuracy is affected greatly by the presence of noise. In this talk we present a novel noise removal and speech enhancement technique based on spiking neural network (SNN) processing of speech data. The SNN has a recurrent lateral inhibitory topology that makes frequency channels compete with one another to remove uncorrelated noise. The SNN can be automatically configured for different acoustic environments and it will be demonstrated how the connectivity results in different time-frequency behaviour. Future directions for further development of this novel approach to noise removal and signal processing will be discussed.
Dr. Cornelius Glackin graduated from the University of Ulster, School of Computing & Intelligent Systems with a MSc in Computing & Intelligent Systems in 2004. Cornelius completed a PhD concerning Spiking Neural Network research at the University of Ulster in 2009. After six years post-doctoral research experience working at the University of Ulster and the University of Hertfordshire, he then moved to industry. Cornelius is currently employed as a Research Scientist at Intelligent Voice Ltd working on machine learning approaches to signal processing, language modelling and speech recognition. Cornelius’ other research interests include: kernel machines, information theory, and robotics.