Effects of noise suppression on spatial release from masking in simulated realistic listening environments

Signal processing methods that eliminate noise from noisy speech signals have the potential to be beneficial in adverse listening environments, particularly for hearing-impaired listeners. However, existing noise suppression techniques have typically been developed to operate on monaural signals. Applying such methods independently to the two ears seems likely to diminish the advantages for speech recognition in noise that are derived from interaural timing and level differences. The present study examines the effect on spatial release from masking of the application of noise suppression based on minimal mean squared error spectral estimation independently to the two ears of normal hearing listeners in different simulated listening environments. These environments are implemented using head-related impulse responses and include an anechoic room, a university meeting room, a school classroom and a cafeteria. Speech reception thresholds (SRTs), defined as the signal-to-noise ratio at which 50% of words are correctly identified, are obtained for IEEE sentences. Target speech is always presented from 0˚ azimuth, while maskers, which include speech shaped noise and interfering speech, are either co-located with the target or spatially separated. Spatial release from masking, i.e., the improvement in SRT with spatial separation of target and masker, will be compared across conditions with and without noise suppression. The data presented will be informative regarding the extent to which non-linear noise suppression processing is detrimental to the effectiveness of binaural spatial cues and will be used in the development of binaural metrics that can predict the intelligibility of processed speech in realistic environments. This, in turn, will contribute to the development of binaural signal processing methods for hearing aids that reduce noise while preserving important spatial cues.