Estimação Robusta do Desvio-Padrão de Ruídos Acústicos Aplicada ao Realce de Sinais de Voz

This paper presents a method of speech enhancement performed in the time domain using a technique robust standard deviation. The technique was evaluated from three experiments, one using only the method and the other two as a proposal for post processing. The techniques used for the initial treatment applied to speech signals corrupted in the second experiment are SS (spectral subtraction) and IMCRA (improved minima controlled recursive averaging). The results presented by the gain of SegSNR indicate that the method provides a gain compared to IMCRA and SS compared individually. Already in experiment post-processing using the method in speech signals processed by SS showed gains, which did not occur as this method was applied to IMCRA.