New method tracks how people listen to a single voice

A new pipeline uses brain signals to estimate how strongly a listener attends to speech.

Absolute auditory attention decoding (aAAD) aims to track how strongly a listener attends to a single speech source. This process uses electroencephalography (EEG).

The aAAD approach lacks a competing speech reference. This makes reliable attention tracking challenging.

The proposed pipeline combines neural envelope tracking (NET) with hidden Markov model (HMM) based temporal inference. NET uses EEG signals and the target speech envelope. HMM is used for temporal inference.

The framework estimates a continuous attention trajectory without explicit attention-state labels. It was evaluated using EEG recordings from 30 adults.

With 1-s analysis windows, non-causal HMM post-processing increased the mean area under the receiver operating characteristic curve (AUC) from 0.547 for raw NET to 0.832. A 5-s window achieves the highest mean AUC of 0.858.

The HMM achieves higher AUC across examined smoothing scales compared to non-causal moving-average (MA) smoothing. This framework provides a label-free approach for estimating temporal fluctuations in auditory attention.

Key facts

  • The framework uses EEG recordings from 30 adults.
  • A 5-s window achieves a mean AUC of 0.858.
  • NET outperforms alpha power and spectral entropy as an observation feature in the unsupervised HMM framework.

All sections

Edition