Work

Exploring sub-band cepstral distances for more robust speaker classification

Year: 2018

Type: conference-paper

Abstract: This paper presents the first of two-part exploration into the potential of parametric cepstral distance (PCD) as a forensic voice comparison feature, based on Japanese vowel data collected from 306 male native speakers under microphone and mobile transmission conditions. The behaviours of PCDs were closely examined by altering sub-band settings, and we found the behaviour of PCDs to correspond well to what is known about formants, which suggests that PCDs are relatable to articulatory gestures. Comparison between sub-band and full-band PCD revealed that limiting the band range to a specific frequency region makes the feature more robust against channel mismatch, encouraging further examination of this potential feature.

Source: ANU Open Research (Australian National University)

Authors: Takashi Osanai, Yuko Kinoshita, Frantz Clermont

Institution: National Research Institute of Police Science

Language: English

Topic: Speech Recognition and Synthesis

Subfield: Artificial Intelligence

Field: Computer Science

Domain: Physical Sciences

Open Access status: green

Funder: Japan Society for the Promotion of Science

Awards: Detection of synthetic voice and character identification for estimation of products., Research on speaker recognition method that is robust to the differences in speaking styles and timing of recording speech using the Standardized-Normalization Transformation

OpenAlex ID: W2953520057

FWCI:

Cites: 9

Cited by: 1

Related to: 10