A new method which will allow more effective voice discrimination has recently been developed by researchers at
Dr Aladdin Ariyaeeinia at the University’s School of Electronic Communication & Electrical Engineering and his team have been conducting research into voice biometrics (speaker recognition) for over 10 years. The process has various potential applications such as verifying individuals’ identities when they try to access cash machines or try to bank or shop online.
The team’s most recent development is a new approach to speaker change detection, a process which captures when speakers change in a given conversational audio stream which could have very useful applications in criminal investigations and in managing audio-visual recordings.
According to Dr Ariyaeeinia, their new approach, which incorporates Gaussian mixture models, is significantly more effective than any current techniques for speaker change detection.
‘We have proved that through the use of this new method, the accuracy of speaker change detection can be enhanced by 30%,’ said Dr Ariyaeeinia.According to Dr Ariyaeeinia, this is achieved through an effective means for tackling the problem posed by phonetic variation in speaker change detection, and by providing the capability for dealing with undesired effects of variations in speech characteristics.