Hanbeot Park

Hanbeot Park, Yunjeong Cho, Hunhee Kim

Restoring speech communication from neural signals is a central goal of brain-computer interface research, yet EEG-based speech reconstruction remains challenging due to limited spatial resolution, susceptibility to noise, and the absence of temporally aligned acoustic targets in imagined speech. In this study, we propose an EEG-to-Voice paradigm that directly reconstructs speech from non-invasive EEG signals without dynamic time warping (DTW) or explicit temporal alignment. The proposed pipeline generates mel-spectrograms from EEG in an open-loop manner using a subject-specific generator, followed by pretrained vocoder and automatic speech recognition (ASR) modules to synthesize speech waveforms and decode text. Separate generators were trained for spoken speech and imagined speech, and transfer learning-based domain adaptation was applied by pretraining on spoken speech and adapting to imagined speech. A minimal language model-based correction module was optionally applied to correct limited ASR errors while preserving semantic structure. The framework was evaluated under 2 s and 4 s speech conditions using acoustic-level metrics (PCC, RMSE, MCD) and linguistic-level metrics (CER, WER). Stable acoustic reconstruction and comparable linguistic accuracy were observed for both spoken speech and imagined speech. While acoustic similarity decreased for longer utterances, text-level decoding performance was largely preserved, and word-position analysis revealed a mild increase in decoding errors toward later parts of sentences. The language model-based correction consistently reduced CER and WER without introducing semantic distortion. These results demonstrate the feasibility of direct, open-loop EEG-to-Voice reconstruction for spoken speech and imagined speech without explicit temporal alignment.

Hanbeot Park, Yunjeong Cho, Hoon-Hee Kim

Handling class imbalance remains a central challenge in machine learning, particularly in pattern recognition tasks where rare but critical events-such as fraudulent transactions or medical anomalies-must be identified accurately. Traditional generative models offer a potential remedy through data augmentation but often treat generation and classification as independent processes, leading to distribution mismatch and limited classifier benefit. To address these shortcomings, we propose Causal Cooperative Networks (CCNETS), a modular learning framework that integrates generation, inference, and reconstruction within a unified causal paradigm. CCNETS comprises three cooperative modules: an Explainer for latent feature abstraction, a Reasoner for label prediction, and a Producer for context-aware data generation. These components interact through a causal feedback loop, where classification results guide targeted sample synthesis. A key innovation, the Zoint mechanism, enables adaptive fusion of latent and observable features, enhancing semantic richness and enabling robust decision-making under uncertainty. We evaluate CCNETS on a real-world credit card fraud detection dataset with extreme imbalance (fraud cases < 0.2%). Across three experimental setups-including synthetic training, amplified generation, and direct classifier comparison-CCNETS outperforms baseline methods, achieving higher F1 scores, precision, and recall. Models trained on CCNETS-generated data also demonstrate superior generalization under limited data conditions. These results establish CCNETS as a scalable, interpretable, and hybrid soft computing framework. By causally aligning synthetic data with classifier objectives, CCNETS advances imbalanced pattern recognition and opens new directions for robust, modular learning in real-world applications.