Realce EMDF e Treinamento em Múltiplas Condições Acústicas para Identificação de Locutor Robusta a Ruídos Não-Estacionários

This paper investigates the multicondition training (MT) technique for improving the speaker identification in non-stationary acoustic noises. The main contribution is the adoption of MT together with a speech enhancement approach based on the empirical mode decomposition. The MT techniques adopted in this work are based on artificial noises with white and colored spectra. Speaker identification experiments are conducted with test utterances corrupted with several acoustic noises and different signal-to-noise ratios. The results show that the MT improves the robustness of the speaker identification task.