Processamento de Sinais Acústicos: Efeito de Distorções Não-Estacionárias na Classificação de Sinais de Voz

Speech classification systems showed to be very interesting for many applications with security needs, such as access control, electronic bank transactions authentication, forensics and biomedical signal processing. Several studies and experiments confirmed that these systems achieve high recognition rates when considering clean or neutral speech signals. However, their performance can be severely degraded when the speech signals are affected by environmental noises (plane, train, car, siren, machine gun) or acoustic emotion (anger, fear, sadness). This paper highlights some solutions proposed in the literature that could improve the robustness of the speech classification systems to these acoustic distortions.