A joint framework for blind prediction of binaural speech intelligibility and perceived listening effort.

Speech perception is strongly affected by noise and reverberation in the listening room, and binaural processing can substantially facilitate speech perception in conditions when target speech and maskers originate from different directions. Most studies and proposed models for predicting spatial unmasking have focused on speech intelligibility. The present study introduces a model framework that predicts both speech intelligibility and perceived listening effort from the same output measure. The framework is based on a combination of a blind binaural processing stage employing a blind equalization cancelation (EC) mechanism, and a blind backend based on phoneme probability classification. Neither frontend nor backend require any additional information, such as the source directions, the signal-to-noise ratio (SNR), or the number of sources, allowing for a fully blind perceptual assessment of binaural input signals consisting of target speech mixed with noise. The model is validated against a recent data set in which speech intelligibility and perceived listening effort were measured for a range of acoustic conditions differing in reverberation and binaural cues [Rennies and Kidd (2018), J. Acoust. Soc. Am. 144, 2147-2159]. Predictions of the proposed model are compared with a non-blind binaural model consisting of a non-blind EC stage and a backend based on the speech intelligibility index. The analyses indicated that all main trends observed in the experiments were correctly predicted by the blind model. The overall proportion of variance explained by the model (R² = 0.94) for speech intelligibility was slightly worse than for the non-blind model (R² = 0.98). For listening effort predictions, both models showed lower prediction accuracy, but still explained significant proportions of the observed variance (R² = 0.88 and R² = 0.71 for the non-blind and blind model, respectively). Closer inspection showed that the differences between data and predictions were largest for binaural conditions at high SNRs, where the perceived listening effort of human listeners tended to be underestimated by the models, specifically by the blind version.

Modeling Binaural Unmasking of Speech Using a Blind Binaural Processing Stage.

The equalization cancellation model is often used to predict the binaural masking level difference. Previously its application to speech in noise has required separate knowledge about the speech and noise signals to maximize the signal-to-noise ratio (SNR). Here, a novel, blind equalization cancellation model is introduced that can use the mixed signals. This approach does not require any assumptions about particular sound source directions. It uses different strategies for positive and negative SNRs, with the switching between the two steered by a blind decision stage utilizing modulation cues. The output of the model is a single-channel signal with enhanced SNR, which we analyzed using the speech intelligibility index to compare speech intelligibility predictions. In a first experiment, the model was tested on experimental data obtained in a scenario with spatially separated target and masker signals. Predicted speech recognition thresholds were in good agreement with measured speech recognition thresholds with a root mean square error less than 1 dB. A second experiment investigated signals at positive SNRs, which was achieved using time compressed and low-pass filtered speech. The results demonstrated that binaural unmasking of speech occurs at positive SNRs and that the modulation-based switching strategy can predict the experimental results.

Modeling Sluggishness in Binaural Unmasking of Speech for Maskers With Time-Varying Interaural Phase Differences.

In studies investigating binaural processing in human listeners, relatively long and task-dependent time constants of a binaural window ranging from 10 ms to 250 ms have been observed. Such time constants are often thought to reflect "binaural sluggishness." In this study, the effect of binaural sluggishness on binaural unmasking of speech in stationary speech-shaped noise is investigated in 10 listeners with normal hearing. In order to design a masking signal with temporally varying binaural cues, the interaural phase difference of the noise was modulated sinusoidally with frequencies ranging from 0.25 Hz to 64 Hz. The lowest, that is the best, speech reception thresholds (SRTs) were observed for the lowest modulation frequency. SRTs increased with increasing modulation frequency up to 4 Hz. For higher modulation frequencies, SRTs remained constant in the range of 1 dB to 1.5 dB below the SRT determined in the diotic situation. The outcome of the experiment was simulated using a short-term binaural speech intelligibility model, which combines an equalization-cancellation (EC) model with the speech intelligibility index. This model segments the incoming signal into 23.2-ms time frames in order to predict release from masking in modulated noises. In order to predict the results from this study, the model required a further time constant applied to the EC mechanism representing binaural sluggishness. The best agreement with perceptual data was achieved using a temporal window of 200 ms in the EC mechanism.