New objective measures for characterisation of noise suppression algorithms

We present two new objective quality measures for the assessment of the performance of noise suppression (NS) algorithms. The signal-to-noise ratio improvement (SNRI) measure attempts to characterise the capability of an NS method to enhance the speech component of a noisy speech signal from an additive background noise. The SNRI measure includes a segmentation of the input speech signal into three power classes, enabling a distinction of the NS effect separately on strong, medium and weak speech. The noise power level reduction (NPLR) measure indicates the impact of an NS method on the power level of the background noise signal in the vicinity of speech activity. The measures were used to produce objective characterisation of AMR/NS (noise suppression for AMR) candidate solutions and are planned to be included in the AMR/NS minimum performance requirement specification. We also present a correlation analysis between the objective measure results and subjective listening test results in conjunction with the GSM EFR speech codec and discuss the relation between the obtained subjective and objective results