Aplicação de estatísticas de formantes e MVKD à comparação forense de locutor

Carregando...
Imagem de Miniatura

Data

Título da Revista

ISSN da Revista

Título de Volume

Editor

Universidade Federal de Minas Gerais

Descrição

Tipo

Artigo de evento

Título alternativo

Primeiro orientador

Membros da banca

Resumo

Forensic speaker comparison (FSC) consists of comparing an unknown audio recording to a known one with the aim of determining whether both recordings come from the same individual. The unknown recording comes from phone lawful interception, which means it is narrowband, GSM-encoded and corrupted by channel noise. This work examines the discriminating power of descriptive statistics computed from formants in short segments of speech. In an attempt to emulate forensic conditions, the recordings were contaminated with six levels of pink noise. Comparisons were performed by a likelihood ratio (LR) framework using the Multivariate Kernel Density (MVKD). The best mean equal error rate (EER) obtained was 12.3% combining formant frequency and band.

Abstract

Assunto

Processamento de sinais - Técnicas digitais, Eletrônica digital

Palavras-chave

Forensic Speaker Comparison, Formant analyze, Multivariate Kernel-Density, Equal Error Rate

Citação

Curso

Endereço externo

https://biblioteca.sbrt.org.br/articlefile/1697.pdf

Avaliação

Revisão

Suplementado Por

Referenciado Por