Decoding the attended speech stream with multi-channel EEG: implications for online, daily-life applications
Bojana Mirkovic1, Stefan Debener, Manuela Jaeger
1Neuropsychology Lab, Department of Psychology, Carl von Ossietzky University of Oldenburg, Ammerländer Heerstr. 114-118, D-26129 Oldenburg, Germany. Cluster of Excellence Hearing4all, Carl von Ossietzky University of Oldenburg, Germany.
Journal of Neural Engineering
|June 3, 2015
Summary
Attended speech decoding using electroencephalography (EEG) is feasible with fewer channels and minimal training data. This advancement brings efficient, real-time brain-computer interfaces for hearing assistance closer to reality.
Area of Science:
- Neuroscience
- Biomedical Engineering
- Signal Processing
Background:
- Temporal envelope driven speech decoding from electroencephalography (EEG) and magnetoencephalography (MEG) can identify attended speech in multi-speaker scenarios.
- Previous studies demonstrated the potential of high-density EEG for this task.
Purpose of the Study:
- Replicate previous high-density EEG findings on attended speech decoding.
- Investigate technical requirements for practical, low-density EEG attended speech decoding.
- Determine minimum training data duration for successful classification.
Main Methods:
- Recorded high-density EEG from 12 participants attending to one of two audiobooks.
- Used an iterative channel elimination procedure to assess channel number requirements.
- Employed chronological cross-validation to determine minimum training data duration.
Main Results:
- Decoder performance remained stable from 96 channels down to 25 channels.
- A subject-independent decoder trained on less than 15 minutes of data outperformed an individually trained decoder.
- Replication confirmed the robustness of the speech decoding method.
Conclusions:
- Efficient low-density EEG online decoding for attended speech is achievable.
- Findings suggest practical applications for hearing assistance technologies.
- The study provides crucial data for developing real-time brain-computer interfaces.
