Robust Speaker Identification Using Matrix Completion Under a Missing Data Imputation Framework

In this work, we examine the problem of noise robust speaker identification under short training sessions restrictions. In order to improve the identification performance, the effects of noise prior to identification must be alleviated. Towards this direction, we exploit matrix completion to recover the unreliable spectrographic data due to noise corruption. This is done by taking advantage of the low rank property of the speech magnitude STFT spectrogram.

In Proc. Workshop on Signal Processing with Adaptive Sparse Structured Representations (SPARS ‘13), Lausanne, Switzerland