i-Vector Transformation Using a Novel Discriminative Denoising Autoencoder for Noise-Robust Speaker Recognition

This paper proposes i-vector transformations using neural networks for achieving noise-robust speaker recognition. A novel discriminative denoising autoencoder (DDAE) is employed on i-vectors to remove additive noise effects. The DDAE is trained to denoise and classify noisy i-vectors simultaneously, making it possible to add discriminability to the denoised i-vectors. Speaker recognition experiments on the NIST SRE 2012 task shows 32% better error performance as compared to a baseline system. Also, our proposed method outperforms such conventional methods as multi-condition training and a basic denoising autoencoder.