Instrumental Quality Predictions and Analysis of Auditory Cues for Algorithms in Modern Headphone Technology

Abstract

Smart headphones or hearables use different types of algorithms such as noise cancelation, feedback suppression, and sound pressure equalization to eliminate undesired sound sources or to achieve acoustical transparency. Such signal processing strategies might alter the spectral composition or interaural differences of the original sound, which might be perceived by listeners as monaural or binaural distortions and thus degrade audio quality. To evaluate the perceptual impact of these distortions, subjective quality ratings can be used, but these are time consuming and costly. Auditory-inspired instrumental quality measures can be applied with less effort and may also be helpful in identifying whether the distortions impair the auditory representation of monaural or binaural cues. Therefore, the goals of this study were (a) to assess the applicability of various monaural and binaural audio quality models to distortions typically occurring in hearables and (b) to examine the effect of those distortions on the auditory representation of spectral, temporal, and binaural cues. Results showed that the signal processing algorithms considered in this study mainly impaired (monaural) spectral cues. Consequently, monaural audio quality models that capture spectral distortions achieved the best prediction performance. A recent audio quality model that predicts monaural and binaural aspects of quality was revised based on parts of the current data involving binaural audio quality aspects, leading to improved overall performance indicated by a mean Pearson linear correlation of 0.89 between obtained and predicted ratings.

Towards a simplified and generalized monaural and binaural auditory model for psychoacoustics and speech intelligibility

“…Hence, a comparison with other state-of-the-art comparison models is given in Section 5.3. However, as the current approach is based on former work [9,[16][17][18]26], it is assumed to generalize well to other unknown data, which makes the model interesting in the context of instrumental (spatial) audio quality predictions. Further, the straightforward processing in the BMFD stage is generally advantageous for real-time applications, e.g., for control of signal processing algorithms in hearing supportive devices or as a hearing-aid processing stage itself.…”

“…Auditory models have been used to explain and analyze monaural and binaural psychoacoustic phenomena (e.g., [5][6][7][8][9]), and as supportive tools offering instrumental assessment of, e.g., speech intelligibility (SI) and audio quality, that are applicable for the development and control of signal processing (e.g., [10][11][12][13][14][15][16]). In such applications, the spectro-temporal composition and interaural differences in the original signal are typically altered [17,18]. Accordingly, monaural cues relevant for, e.g., spectral and temporal masking [19,20], and binaural cues relevant for, e.g., sound-source location and apparent source width [15] might be affected.…”

“…Here we suggest and examine a combined monaural and binaural model in a variety of "benchmark" psychoacoustic and speech-intelligibility experiments. The combined approach uses the monaural front end and back end of the generalized power spectrum model (GPSM; [26]) which has been successfully applied to monaural psychoacoustics, speech intelligibility, and audio quality assessment [9,[16][17][18]26]. A binaural processing stage with five fixed (nonadaptive) output channels is suggested, referred to as binaural matrix feature decoder (BMFD).…”

Binaural detection thresholds and audio quality of speech and music signals in complex acoustic environments

“…No large effect of reverberation time was found, suggesting that the use of only a single or few reverberation times might be sufficient for the audio quality assessment of such devices. Only a few approaches, e.g., Cauchi et al (2019) , and Biberger et al (2021) considered aspects of reverberation affecting quality predictions. Biberger et al (2021) found monaural spectral cues, capturing spectral coloration distortions of hearing devices aiming at acoustically transparency, to be more reliable for quality predictions in reverberation than cues based on the temporal fine structure or cepstrum correlation.…”

“…Some auditory perception models have been applied to isolated aspects of CAEs. One example is the (monaural) Generalized Power Spectrum Model (GPSM), which has been applied to psychoacoustic masking with simplified psychoacoustic stimuli ( Biberger and Ewert, 2016 , 2017 ) as well as to audio quality for various distortions in anechoic and echoic conditions without maskers ( Biberger et al, 2018 , 2021 ).…”