A Cutting-Edge Feature Extraction Approach for Speaker Recognition Leveraging Optimized Variance Spectral Flux and Daubechies Wavelet

Automatic Speaker Recognition (ASR) is a crucial application in the realm of speech processing, with Artificial Intelligence (AI) being extensively employed in areas such as authentication, surveillance, forensics, and security. The cornerstone processes of these applications encompass feature matching, feature extraction, and performance evaluation. However, the present speaker identification and verification techniques are not without their flaws, including vulnerability to distortion resulting from noise and the ability to mimic signals via voice recording devices. Given these challenges, there's a pressing need for a fresh feature extraction technique that offers robust speaker identification using an enhanced spectrogram. This paper addresses this need by proposing an innovative and efficient feature extraction methodology, christened "Optimized Variance Spectral Flux (OVSF)". This potent technique, based on the Daubechies 40 wavelet and power spectrum of a signal, facilitates the extraction of unique features of the speaker. For the feature matching phase in speaker recognition, the characteristics of different speakers are compared by applying the time-honored Bayesian information criterion distance metric. The proposed system's effectiveness is assessed through a series of metrics including Receiver Operating Characteristics (ROC), detection error trade-off curves, the Equal Error Rate (EER), and the Area under the Curve (AUC). The experimental results yield an AUC and EER for the proposed method of 94.38 and 10.3564, respectively, indicating a higher accuracy than the mel-frequency cepstral coefficient technique.