We are pleased to share that Asma Sbaih, together with Jorge García-Gutiérrez, Megha Bhushan and David Benavides, has published a new article entitled “Deep Learning Methods for EEG-Based Speech Classification and Decoding: A PRISMA Review”, now available in open access in Computer Speech & Language (Elsevier).
The article presents a systematic review, conducted following the PRISMA 2020 guidelines, of deep learning methods applied to speech-related EEG (electroencephalography) and intracranial EEG (iEEG) processing, covering studies published between 2018 and 2025. After searching across Scopus, IEEE Xplore, ScienceDirect, Web of Science and PubMed, the authors identified 1,148 records, of which 80 peer-reviewed studies were included following duplicate removal, screening and eligibility assessment.
The review organizes the literature by task type (speech classification, spectrogram reconstruction and speech synthesis), neural signal type (non-invasive EEG versus invasive ECoG and sEEG) and model architecture. Among its findings, deep learning models clearly outperform traditional methods in speech classification, while spectrogram reconstruction and speech synthesis remain more challenging, especially for non-invasive EEG; invasive recordings consistently achieve better results for these latter tasks. The review also highlights that few studies currently address real-time feasibility or cross-subject generalization, two key barriers to clinical translation.
This work contributes a structured, task-oriented overview of deep learning–based neural speech decoding and identifies the methodological and translational challenges that must be addressed to enable robust, real-time, clinically viable speech neuroprosthetic systems.
The full article is available open access at the following link: https://doi.org/10.1016/j.csl.2026.102020
We congratulate Asma Sbaih and her co-authors on this excellent contribution to the field of EEG-based speech decoding.