AMADEUSAMADEUS
A factor important to the success of a variety of ubiquitous computing applications is that of identification of users so that each individual is provided with an appropriate level of authorisation. This can be for a variety of purposes both at home and work environments. Examples are controlling appliances, physical access to restricted areas, access to electronic information, launching software programmes, access to data over the Internet and online financial transactions. A difficulty with the conventional means of identification is that they are not designed for use in smart environments, and can be easily compromised. In view of these, it appears that the required optimal usability and reliability in determining the identities of users may only be achieved through the deployment of user-friendly biometrics. An identification method in this category is speaker recognition (voiceprint). It is user friendly, and also non-contact. A main component of any speaker recognition system, and also any speech-based interactive system in general, is the speech feature extraction engine (FEE).
The aim of this project is to develop hardware embodiments of algorithms for effective audio feature extraction in order to create a digital hardware engine (an IP-Core) for use in SOC integration based on FPGA. The feature extraction engine (FEE) will present a standardised parametric model for the representation of voice characteristics that subsequent classification system components can utilise to achieve final signature matching. This will ensure that designs utilising speech feature parameters can maintain compatibility even though future FEE's might be improved internally to achieve better performance and reliability. The effectiveness of voice classification, however, can be significantly affected by degradation in speech, mainly due to additive environmental noise. This project seeks to address the audio pre-conditioning (noise reduction) and feature extraction issues primarily, with a demonstrator outcome based upon the requirements of a home environments scenario.
©2003 Dept. of Computer Science, The University of York.