Employment of voicing information of speech spectra for noise-robust speaker identification

摘要

This paper presents a novel method for voicing information estimation of individual frequency-regions of speech spectra and its employment in a text-independent speaker identification system. The voicing information is incorporated to the system in a form of a mask in a marginalization-based missing-feature model. Experiments were performed on speech data from the TIMIT database corrupted by stationary and real-world noises. The obtained results show that using the proposed voicing estimation method provides performance close to using oracle voicing information. The combination of the voicing information mask with a noise-estimate mask showed further improvement in the identification accuracy and achieved performance close to the oracle mask obtained using full a-priori knowledge of noise.