Transcranial alternating current stimulation (tACS) can non-invasively modulate neuronal activity in humans and influence auditory perception. Recent studies have shown that tACS with the alternating current that follows the envelope of a speech signal can modulate the comprehension of this voice in background noise. However, how exactly tACS influences cortical activity and affects speech comprehension remains poorly understood. Here, we present a computational model for speech coding in a spiking neural network and employ it to investigate the effects of tACS on the neural encoding of speech in noise.

Based on previous work, we established a spiking neural network model for speech encoding. The model consisted of two coupled neuronal populations generating self-sustained oscillations in theta (4-8 Hz) and gamma (above 25 Hz) frequency ranges. The theta-generating module was designed to track the onsets of syllables and parse the faster gamma activity into smaller segments. To quantify the speech-in-noise encoding performance of the model, we simulated the encoding of spoken sentences in different levels of background noise and used the obtained spiking outputs to decode syllable identities. Furthermore, we subjected the model to a range of speech-inspired tACS waveforms and investigated their effects on the speech-in-noise encoding.

Both syllable tracking and decoding deteriorated in a sigmoidal fashion with increasing levels of background noise. The model performance in the presence of noise resembled typical human comprehension in the analogous speech-in-noise listening task. The simulated tACS interventions yielded phase- and time-dependent modulation of the speech-in-noise encoding in the model, similar to the effects of tACS on the speech-in-noise comprehension. The greatest facilitation of syllable tracking in the model was observed for tACS preceding the speech signal by 50-100 ms. Notably, this latency range corresponds to the typical neural delay associated with cortical auditory processing in humans and thus may explain why tACS applied without additional latency can influence comprehension. Stimulation waveforms filtered in the theta-band frequency range impacted the speech-in-noise encoding in the model the most, which reflected the effects of tACS in humans.

The proposed model provides biophysically-grounded estimates of neural encoding of speech in noise, suitable for studying the relationship between cortical processing of speech and comprehension. The effects of tACS interventions predicted by the model agree with recent experimental findings and shed light on neural mechanisms through which the behavioural effects may arise.