Feature Level Compensation for Robust Speaker Identification in Mismatched Conditions

In this paper, robust front end features are proposed for improvement in speaker identification (SI) performance by considering the factors of real world situations, like mismatch between training and testing conditions. The most commonly used MFCC features are very much sensitive to effects such as channel and environment mismatch. Characteristics of speech gets changed with room acoustics, channel and microphone as well as background noise, which adversely affects the performance of the SI system. To make the front end features robust, asymmetric hamming-cosine taper is used, which gives better spectral estimation and reduces the interfering band limited noise. To incorporate time varying information, second order derivatives of cepstral coefficients are concatenated to MFCC features. Convolutional errors are minimized by using cepstral mean normalization (CMN) and compensation to additive noise is achieved by magnitude spectral subtraction (MSS). The performance of closed-set text independent speaker identification system is evaluated under different train and test conditions such as sensor (microphone) mismatch, speaking style mismatch, language mismatch and environment mismatch using IIT-G multi-variability speech database developed for speaker recognition purpose. The experimental results shows that, these modified front end features outperform conventional baseline MFCC features in mismatched conditions especially for sensor and environment mismatch. It is observed that changes in speaking style and language mismatch affects less on performance accuracy. Keywords—Text Independent SpeakerIdentification, MFCC,Cepstral Mean Normalization, Magnitude Spectral Subtraction