Low Complexity Speaker Identification Approach

Biometric authentication presents an increasing interest in the last years. Among the usual biometric features employed in the authentication systems speech plays an important role and is well accepted by the users. Speech as biometric verifier has some important advantages in use, respectively: the low price for the sensors (microphone), the easy way (noninvasive) of samples acquisition, the large amount of available data for training and testing and the ability to provide real time processing. Our approach employs only features extracted in the time domain, so two orders smaller computational requirements are implied than in the case of usual frequency domain processing. Due to this quality a low complexity and low cost implementation may be done using an ATMEL microcontroller platform. The system is based on the TESPAR coding method for the feature extraction from speech signal allowing a real time processing on a medium power computation platform. The system performances proves the discrimination capability of the TESPAR method even when using only the distances for the classification task and encourages the improvement of the system for low cost applications. The method employment may be extended to other applications like biomedical waveform diagnostics (EEG, ECG) or industrial system monitoring. [PUBLICATION ABSTRACT]