A speaker recognition method based on GMM using non-negative matrix factorization

Speaker recognition enhances communication efficiency based on dialogue. In this paper, we propose a speaker recognition method based on feature fusion with nonnegative matrix decomposition (NMF). It uses NMF to decompose the speaker's speech spectrogram, extract features from the decomposed basis matrix, and perform feature fusion with Mel-frequency cepstral coefficients (MFCC). Experimental results show that the fused features improve speaker recognition accuracy in low signal-to-noise (SNR) environment. Comparison with other speaker recognition methods using MFCC features.