Target speaker identification is essential for speech enhancement algorithms in assistivedevices aimed toward helping the hearing impaired. Several recent studies have reportedthat target speaker identification is possible through electroencephalography (EEG)recordings. If the EEG system could be reduced to acceptable size while retainingthe signal quality, hearing aids could benefit from the integration with concealedEEG. To compare the performance of a multichannel around-the-ear EEG systemwith high-density cap EEG recordings an envelope tracking algorithm was appliedin a competitive speaker paradigm. The data from 20 normal hearing listeners wereconcurrently collected from the traditional state-of-the-art laboratory wired EEG systemand a wireless mobile EEG system with two bilaterally-placed around-the-ear electrodearrays (cEEGrids). The results show that the cEEGrid ear-EEG technology capturedneural signals that allowed the identification of the attended speaker above chance-level,with 69.3% accuracy, while cap-EEG signals resulted in the accuracy of 84.8%. Furtheranalyses investigated the influence of ear-EEG signal quality and revealed that theenvelope tracking procedure was unaffected by variability in channel impedances. Weconclude that the quality of concealed ear-EEG recordings as acquired with the cEEGridarray has potential to be used in the brain-computer interface steering of hearing aids.
展开▼
