It might be far more effective to capture subvocalisations using surface electrodes and then running those signals through machine learning: https://annals-csis

It might be far more effective to capture subvocalisations using surface electrodes and then running those signals through machine learning: https://annals-csis.org/Volume_11/drp/pdf/153.pdf https://annals-csis.org/Volume_11/drp/pdf/153.pdf