BCI for Auditory Attention: Can Neurofeedback Train Your Brain to Ignore Distractions?

Auditory attention decoding (AAD) algorithms have made progress in identifying which speaker a listener is following. BCI for auditory attention is not only about decoding where a listener focuses. It may also help train the brain to focus better. But most work in this area has focused on decoding alone.

A recent study asked a different question: can real-time neurofeedback improve focus? The authors used a closed-loop EEG system to test whether a single session could change how the brain processes competing speech.

Experiment:

Participants sat in a sound-treated booth between two loudspeakers placed at ±45 degrees. Each loudspeaker presented a different male voice reading an audiobook. The task was to maintain attention on one talker while ignoring the other.

The session lasted one hour and consisted of two phases.

During the first phase, the system collected data from 10 one-minute trials. Those data were then used to train an individualized decoder for each participant. Attention alternated between the left and right talker across trials. This gave the decoder balanced data from both conditions. The short calibration phase is important for practical use. BCI for auditory attention should work in realistic settings – without lengthy preparation.

The second phase consisted of 50 one-minute trials. Each participant attended to one assigned talker, balanced across the group. Here, the neurofeedback loop became active. As decoding accuracy improved, the system reduced the unattended talker’s volume. When attention drifted, the distractor came back up.

So how might attention improve?

The authors did not only ask whether performance improved. They also asked how it improved in the brain.. They proposed three possible neural mechanisms:

Hypothesis 1: attention improves because the brain tracks the attended speaker more strongly over time.

Hypothesis 2: attention improves because the brain suppresses its tracking of the ignored speaker.

Hypothesis 3: both mechanisms occur together.

This distinction matters. Better listening can mean different things in the brain. A listener may improve by strengthening the target voice, suppressing the distractor, or both.

Recording and Decoding EEG Data

Researchers used mbt’s portable EEG system for EEG collection.

To decode attention, the authors used a temporal response function (TRF) based approach.

A linear decoder was trained to predict the envelope of the attended talker from EEG data. Once applied to a new segment of EEG data, the decoder produces a predicted attended talker envelope. Which is then correlated with the actual envelopes of both talkers in the scene:

The difference between these two values reflects the strength of neural tracking for the attended talker:

Decoder accuracy is then defined as the fraction of time this difference is greater than zero:

The system updated every 500 ms. Better decoding led to stronger attenuation of the distractor. The unattended talker could be reduced from 0 to −10 dB SPL. The attended talker remained constant. This created a real-time feedback loop between brain activity and the acoustic environment.

Can neurofeedback improve attention in noisy environments?

The decoder performed reasonably well despite the short calibration period. Mean calibration accuracy was 60.8%. Participants who performed better during calibration also tended to perform better later in the session (ρ = 0.59, p = 0.004). This suggests that the system captured meaningful individual differences in auditory attention.

The answer to the question above appears to be yes – but not in the way one might expect. Neurofeedback didn’t improve attention by strengthening neural tracking of the attended talker. Instead, the effect appeared on the distractor side. During feedback-on trials, neural tracking of the unattended talker decreased from the first half of session to the second. This was reflected in significant decreases in both the unattended decoder correlation metric (p = 0.02, Cohen’s d = −1.29) and unattended decoding accuracy (p = 0.01, Cohen’s d = −1.56).

That distinction matters. The brain did not seem to improve by “turning up” the voice of interest. It improved more by turning down the competing one. In the context of BCI for auditory attention, this is an important insight.

This effect was absent in trials without feedback. That suggests the change was related to the neurofeedback itself, not just repetition over time. The authors also found no difference in comprehension accuracy between the first and second halves of the session. They found no difference in unattended talker presentation level across the session either.

Together, these control results make it less likely that the effect was caused by changing engagement or by changes in the stimulus itself.

This may also reflect the design of the feedback itself. Because the system rewarded better decoding by attenuating the distractor. The paradigm naturally encouraged suppression. A different feedback design might have produced a different neural outcome.

Why this matters

This study does more than show real-time decoding. The paper suggests that BCI for auditory attention may help shape attention itself. In a single session, neurofeedback reduced neural tracking of distracting speech. This points to a possible future role for neurofeedback-based auditory training.

Of course, this is still a proof-of-concept study. Still, the idea is compelling. BCI for auditory attention systems may not only decode what we want to hear. They could also help train the brain to ignore what does not matter.

You can read the original study by Stephanie Haro et al. linked here.