Speaker recognition systems can typically attain high performance in ideal conditions. However, significant degradations in accuracy are found in channel-mismatched scenarios. Non-stationary environmental noises and their variations are listed at the top of speaker recognition challenges. Gammtone frequency cepstral coefficient method (GFCC) has been developed to improve the robustness of speaker recognition. This paper presents systematic comparisons between performance of GFCC and conventional MFCC-based speaker verification systems with a purposely collected noisy speech data set. Furthermore, the current work extends the experiments to include investigations into language independency features in recognition phases. The results show that GFCC has better verification performance in noisy environments than MFCC. However, the GFCC shows a higher sensitivity to language mismatch between enrollment and recognition phase.
Click to purchase paper as a non-member or login as an AES member. If your company or school subscribes to the E-Library then switch to the institutional version. If you are not an AES member and would like to subscribe to the E-Library then Join the AES!
This paper costs $33 for non-members and is free for AES members and E-Library subscribers.