Modified MFCCs for robust speaker recognition

Abstract

Mel-scale frequency cepstrum coefficienls (MFCCs) are commonly used katues in speaker recognition systems, but MFCC values are not very robust in the presence of noise. thus, the modified MFCCs (named as SMN-CMN MFCC) based on the general noisy speech model is proposed in this paper, which uses spectrum mean normalization (SMN) to suppress the additive noise, and uses cepstral mean normalization (CMN) to remove the effect of convolutional noise. Theoretical analyses show that the combination of SMN and CMN can inhibit additive and convolutional noise at the same time. To verify the performance of the SMN-CMN MFCC, \\e have conducted some speaker recognition tesls. With the same convolutional noise component, the additive white noise experimenls and the additive factory noise experimenls show that SMN-CMN-MFCC provides 10.5% and 9.6% relative improvement than the conventional MFCC and AMFCC features, respectively.

Robust Feature Extraction Using Temporal Context Averaging for Speaker Identification in Diverse Acoustic Environments

“…For parts of the signal where the number of remaining frames is less than the set context window size N , the context window size N i is adjusted, as specified in (12), to ensure that all frames are included in the analysis. Following this adjustment, the Feature Averaging process computes the average for each coefficient by combining the corresponding coefficients from all frames within the window, as detailed in (13). This averaging operation results in TCEF vectors that provide a richer and more robust representation of the conventional features.…”

“…Additionally, there have been efforts to adapt conventional feature extraction methods to improve their robustness against environmental noise and reverberation. Modifications have been proposed for MFCC [13], GTCC [37], and PNCC [22,47]. These enhanced feature variants introduce additional complexity to the feature extraction process by incorporating extra computational operations, such as adaptive noise compensation or non-linear transformation techniques, resulting in increased computational demands compared to the original feature computation.…”