# CIPHER: Conformer-based Inference of Phonemes from High-density EEG

Source: [CIPHER: Conformer-based Inference of Phonemes from High-density EEG](<https://www.sota2.com/research/paper/69d33e0d7d0172140ec736c7>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-04-07T05:00:04.607Z

Authors: Varshith Madishetty

Published: 2026-03-22T12:14:47.000Z

## Abstract

Decoding speech information from scalp EEG remains difficult due to low SNR and spatial blurring. We present CIPHER (Conformer-based Inference of Phonemes from High-density EEG Representations), a dual-pathway model using (i) ERP features and (ii) broadband DDA coefficients. On OpenNeuro ds006104 (24 participants, two studies with concurrent TMS), binary articulatory tasks reach near-ceiling performance but are highly confound-vulnerable (acoustic onset separability and TMS-target blocking). On the primary 11-class CVC phoneme task under full Study 2 LOSO (16 held-out subjects), performance is substantially lower (real-word WER: ERP 0.671 +/- 0.080, DDA 0.688 +/- 0.096, indicating limited fine-grained discriminability. We therefore position this work as a benchmark and feature-comparison study rather than an EEG-to-text system, and we constrain neural-representation claims to confound-controlled evidence.

## Recorded benchmark results

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

- [Phoneme-level speech decoding on Study 2 CVC Triphones Real Words (LOSO 16 folds)](<https://www.sota2.com/research/sota/phoneme-level-speech-decoding-on-study-2-cvc-triphones-real-words-loso-16-folds.md>): Phoneme-level speech decoding; Study 2 CVC Triphones Real Words 16 folds (LOSO); WER: 0.671.

- [Phoneme-level speech decoding on Study 2 CVC Triphones (Pseudowords) 16 folds (LOSO)](<https://www.sota2.com/research/sota/phoneme-level-speech-decoding-on-study-2-cvc-triphones-pseudowords-16-folds-loso.md>): Phoneme-level speech decoding; Study 2 CVC Triphones (Pseudowords) 16 folds (LOSO); WER: 77.2.

- [Binary Articulatory Feature Classification on Study 2 (val)](<https://www.sota2.com/research/sota/binary-articulatory-feature-classification-on-study-2-val.md>): Binary articulatory feature classification; Study 2 (val); Top-1 Accuracy: 100.

- [Binary Articulatory Feature Classification on Study 1 Cross-Study (test)](<https://www.sota2.com/research/sota/binary-articulatory-feature-classification-on-study-1-cross-study-test.md>): Binary articulatory feature classification; Study 1 Cross-Study (test); Top-1 Accuracy: 100.

- [Place of Articulation Classification on study2 matched (val)](<https://www.sota2.com/research/sota/place-of-articulation-classification-on-study2-matched-val.md>): Place of Articulation Classification; study2 matched (val); ERP Accuracy: 57.3.

- [Manner of Articulation Classification on study2 matched condition (val)](<https://www.sota2.com/research/sota/manner-of-articulation-classification-on-study2-matched-condition-val.md>): Manner of Articulation Classification; study2 matched condition (val); ERP Accuracy: 85.2.

- [Phoneme Identity Classification on study2 matched (val)](<https://www.sota2.com/research/sota/phoneme-identity-classification-on-study2-matched-val.md>): Phoneme Identity Classification; study2 matched (val); ERP Accuracy: 15.5.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/2604.02362>)

[PDF](<https://arxiv.org/pdf/2604.02362.pdf>)