Third International Workshop on Models and Analysis of Vocal Emissions for Biomedical Applications (MAVEBA 2003)

Florence, Italy
December 10-12, 2003

HMM Modelling of Additive Noise in the Western Languages Context

C. S. Lima (1), J. F. Oliveira (2)

(1) Department of Industrial Electronics, Universidade do Minho, Guimarães, Portugal
(2) Department of Electrical Engineering, Instituto Politécnico de Leiria, Leiria, Portugal

This paper is concerned to the noisy speech HMM modelling when the noise is additive, speech independent and the spectral analysis is based on subbands. The internal distributions of the noisy speech HMM's were derived when Gaussian mixture density distributions for clean speech HMM modelling are used, and the noise is normally distributed and additive in the time domain. In these circumstances it is shown that the HMM noisy speech distributions are not Gaussian, however, fitting these distributions as a Gaussian mixture, only a little bit of loss in performance was obtained at very low signal to noise ratios, when compared with the case where the real distributions were computed using Monte Carlo methods.

Full Paper (reprinted with permission from Firenze University Press)

Bibliographic reference.  Lima, C. S. / Oliveira, J. F. (2003): "HMM modelling of additive noise in the western languages context", In MAVEBA-2003, 65-68.