Preference and similarity of combined noise reduction and dynamic range compression in hearing-impaired listeners

Single-channel noise reduction (SCNR) and wide dynamic range compression (WDRC) are used in hearing aids to suppress background noise and enhance speech audibility. However, these processing stages can have opposing effects as SCNR may unintentionally reduce soft speech components, whereas WDRC can amplify SCNR artifacts and reduce signal-to-noise ratio. This study presents a behavioral evaluation of WDRC configurations with and without a deep neural network (DNN)-based SCNR. The WDRC configurations encompassed fast- and slow-acting compression and a scene-aware compressor that switched between fast and slow depending on the speech presence. An "ideal" reference system was included and applied fast-acting compression to the target signal while attenuating noise and reverberation. The systems were evaluated using pairwise comparison tasks to evaluate preference and similarity for speech in noise by 16 participants with hearing impairment. The results indicated that the DNN-based SCNR stage improved the preference for all compression systems. The ideal system was the most preferred, followed by scene-aware- and slow-acting-compression, both with SCNR, whose preference scores were close to the ideal system. The results suggest that effective compression can be applied to the target speech signal through either ideal compression or scene-aware compression without compromising quality, as reflected by the preference judgments.