2013 RISP International Workshop on Nonlinear Circuits, Communications and Signal Processing
Volume, Number, Page
Published date
2013
Publisher
Japanese:
English:
Conference name
Japanese:
English:
2013 RISP International Workshop on Nonlinear Circuits, Communications and Signal Processing
Conference site
Japanese:
English:
The island of Hawaii
File
Abstract
This paper proposed a novel and robust voice activity detection
(VAD) algorithm utilizing long-term spectral flatness
measure (LSFM) which is capable of working at very low
signal-to-noise ratios (less than 10 dB). The proposed algorithm
was evaluated under eleven types of noises from
NOISEX-92 and five types of SNR in core TIMIT TEST corpus.
Comparisons with three modern standardized algorithms
(ETSI AMR option 1&2 and ITU-T G.729 Annex B) demonstrate
that our proposed LSFM-based VAD scheme achieved
highest speech hit rate among the four algorithms while keeping
good hit rate balance between speech hit rate and nonspeech
hit rate.