Three methods of estimation of the a priori signal-to-noise ratio (SNR) in noise reduction algorithms were compared. It is shown that "decision-directed" method ensures the best sound quality of the speech signal. "Rough" estimate was the best for use in automatic speech recognition for the overall SNR > 15 dB. The maximum likelihood estimation occupies an intermediate position between estimates by "decision-directed" and “rough” estimator.
noise interference; noise reduction algorithm; a priori signal-to-noise ratio; quality indicator; speech signal
"Otsenyvanye apryornoho otnoshenyia syhnal-shum v alhorytmakh shumopodavlenyia" [Assessment of a priori signal-to-noise ratio in noise reduction algorithms],
Information Processing Systems,