Subspace modeling of prosodic features for speaker verification

Subspace modeling of prosodic features for speaker verification

Dizertace

The thesis investigates into speaker verification by means of prosodic features. This includesan appropriate representation of speech by measurements of pitch, energy and durationof speech sounds. Two diverse parameterization methods are investigated: the firstleads to a low-dimensional well-defined set, the second to a large-scale set of heterogeneousprosodic features. The first part of this work concentrates on the development of so calledprosodic contour features. Different modeling techniques are developed and investigated,with a special focus on subspace modeling. The second part focuses on a novel subspacemodeling technique for the heterogeneous large-scale prosodic features. The modelis theoretically derived and experimentally evaluated on official NIST Speaker RecognitionEvaluation tasks. Huge improvements over the current state-of-the-art in prosodic speakerverification were obtained. Eventually, a novel fusion method is presented to elegantlycombine the two diverse prosodic systems. This technique can also be used to fuse thehigher-level systems with a high-performing cepstral system, leading to further significantimprovements.

The thesis investigates into speaker verification by means of prosodic features. This includesan appropriate representation of speech by measurements of pitch, energy and durationof speech sounds. Two diverse parameterization methods are investigated: the firstleads to a low-dimensional well-defined set, the second to a large-scale set of heterogeneousprosodic features. The first part of this work concentrates on the development of so calledprosodic contour features. Different modeling techniques are developed and investigated,with a special focus on subspace modeling. The second part focuses on a novel subspacemodeling technique for the heterogeneous large-scale prosodic features. The modelis theoretically derived and experimentally evaluated on official NIST Speaker RecognitionEvaluation tasks. Huge improvements over the current state-of-the-art in prosodic speakerverification were obtained. Eventually, a novel fusion method is presented to elegantlycombine the two diverse prosodic systems. This technique can also be used to fuse thehigher-level systems with a high-performing cepstral system, leading to further significantimprovements.

speaker verification, prosody

speaker verification, prosody

KOCKMANN, M.

21.05.2012

Brno

122

http://www.fit.vutbr.cz/research/groups/speech/publi/2012/phdthesis_kockmann.pdf