Torsten Dau

Environment-aware ideal binary mask estimation using monaural cues

We present a monaural approach to speech segregation that estimates the ideal binary mask (IBM) by combining amplitude modulation spectrogram (AMS) features, pitch-based features and speech presence probability (SPP) features derived from noise statistics. To maintain a high mask estimation accuracy in the presence of various background noises, the system employs environment-specific segregation models and automatically selects the appropriate model for a given input signal. Furthermore, instead of classifying each time-frequency (T-F) unit independently, the a posteriori probabilities of speech and noise presence are evaluated by considering adjacent T-F units. The proposed system achieves high classification accuracy.

Predicting speech intelligibility based on the modulation spectrum and modulation frequency selectivity

Speech intelligibility depends on factors related to the auditory processes involved in sound perception as well as on the acoustic properties of the sound entering the ear. A clear understanding of speech perception in complex acoustic conditions remains a challenge. Here, a computational modeling framework is presented that attempts to predict the speech intelligibility obtained by normal-hearing and hearing-impaired listeners in various adverse conditions. The model combines the concept of envelope frequency selectivity in the auditory processing of the sound with a decision metric that is based either on the signal-to-noise envelope power ratio or a correlation measure. The proposed model is able to account for the effects of stationary background noise, reverberation, nonlinear distortions and noise reduction processing on speech intelligibility. However, due to its simplified auditory preprocessing stages, the model fails to account for the consequences of individual hearing loss on intelligibility. To address this, physiologically inspired extensions of the auditory preprocessing in the model are combined with the modulation-frequency selective processing and the back-end processing that have been successful in the conditions tested with normal-hearing listeners. The goal is to disentangle the consequences of different types of hearing deficits on speech intelligibility in a given acoustic scenario.Speech intelligibility depends on factors related to the auditory processes involved in sound perception as well as on the acoustic properties of the sound entering the ear. A clear understanding of speech perception in complex acoustic conditions remains a challenge. Here, a computational modeling framework is presented that attempts to predict the speech intelligibility obtained by normal-hearing and hearing-impaired listeners in various adverse conditions. The model combines the concept of envelope frequency selectivity in the auditory processing of the sound with a decision metric that is based either on the signal-to-noise envelope power ratio or a correlation measure. The proposed model is able to account for the effects of stationary background noise, reverberation, nonlinear distortions and noise reduction processing on speech intelligibility. However, due to its simplified auditory preprocessing stages, the model fails to account for the consequences of individual hearing loss on intelligibility....

Towards cognitive control of hearing instruments using EEG measures of selective attention

Single-trial EEG measures of selective auditory attention have recently suggested the perspective of decoding who a listener is focusing on in multi-talker scenarios. Here, we report results from work within the COCOHA (Cognitive Control of a Hearing Aid) project investigating the possibility of integrating EEG into neuro-steered hearing instruments. Our EEG decoding strategy relies on measuring cortical activity entrained to envelope fluctuations in the attended speech signal. Currently, a major challenge has been to obtain robust EEG measures of selective attention in older hearing-impaired (HI) listeners. We report our recent COCOHA attempts to decode selective attention from the EEG of hearing-impaired (HI) listeners. Aided HI listeners and age-matched normal-hearing controls were presented with competing talkers at 0 dB target-to-masker ratio and instructed to attend to one talker. We show that single-trial decoding accuracies similar to those reported for younger listeners can be obtained with both ...