A Novel Scoring Method Based on Distance Calculation for Similarity Measurement in Text-Independent Speaker Verification
Soufiane Hourri, Jamal Kharroubi · Procedia Computer Science · 2019
Nowadays, stochastic models are state-of-the-art for text-independent speaker verification. However, they are costly in terms of time-consuming and may need much more data in the training phase. This paper proposes a novel scoring method based on distance calculation for similarity measurement in text-independent speaker verification. The basic idea of our approach aims to propose a new similarity measurement method using, directly, the speaker’s feature vectors (MFCC), in order to preserve and take advantage of the speaker’s specific features. Experiments on two open source speaker recognition corpora confirm our idea. Results demonstrate that our approach largely outperforms state-of-the-art approaches, GMM-UBM and i -vector/PLDA.