Speaker Verification: A Raw Waveform Approach for Text Independent using CNN

Recognizing the speaker’s voice is the goal of speaker recognition. In general, speaker identification and verification fall under speaker recognition. Numerous deep learning-based architectures for speaker recognition have been suggested, as evidenced by the literature. The accuracy of recognition models has significantly improved with the use of i- vector or x-vector feature extraction models, yet, the architecture appears to be very complex. A scalable speaker recognition system requires lightweight models with adequate accuracy to be implemented for various security applications. The suggested framework goes in this direction by utilizing Convolutional Neural Network architecture, which is trained and tested on Libri speech data which includes short utterances of 7000 speakers. The suggested framework is trained by extracting MFCC characteristics from the provided utterance. Comparing the architecture to other cutting-edge methods, it appears to be lightweight with reasonable accuracy.