EMPLOYMENT OF VOICING INFORMATION OF SPEECH SPECTRA FORNOISE-ROBUST SPEAKER IDENTIFICATION

摘要

This paper presents a novel method for voicing informationrnestimation of individual frequency-regions of speechrnspectra and its employment in a text-independent speakerrnidentification system. The voicing information is incorporatedrnto the system in a form of a mask in a marginalizationbasedrnmissing-feature model. Experiments were performedrnon speech data from the TIMIT database corrupted by stationaryrnand real-world noises. The obtained results showrnthat using the proposed voicing estimation method providesrnperformance close to using oracle voicing information. Therncombination of the voicing information mask with a noiseestimaternmask showed further improvement in the identificationrnaccuracy and achieved performance close to the oraclernmask obtained using full a-priori knowledge of noise.