Noninvasive Speech Restoration After Total Laryngectomy Using Surface Electromyography: Pilot Study

To demonstrate the feasibility of restoring intelligible, own-voice speech after total laryngectomy using noninvasive surface electromyography.

One post-laryngectomy patient and one healthy reference speaker silently articulated a closed-vocabulary corpus of 499 dates and times while surface electromyographic (sEMG) signals were recorded from a 20-electrode face, jaw, and neck montage. A decoder mapped sEMG into speaker-independent speech tokens, which a public token-to-speech stack converted into audible speech; for the patient, the output was re-rendered in his own pre-surgical voice using approximately 8 min of audio recorded before surgery. A per-subject procedure identified a reduced electrode subset, and performance was evaluated by five human listeners.

The system produced intelligible speech in the patient's own pre-surgical voice. Mean word error rate by human listeners was 20.2% (95% CI 11.0%-31.5%) for the patient and 14.9% (6.8%-24.9%) for the reference speaker. Electrode-selection analysis indicated that roughly half of the recorded electrodes captured 95% of the multi-channel signal (9 of 20 for the patient, 11 of 18 for the reference), supporting more compact hardware. Single-trial recordings showed coordinated muscle activity tracking each silently articulated word despite the absence of phonation.

Noninvasive sEMG-based speech restoration is feasible after total laryngectomy. By decoding muscle activity into a speaker-independent representation, the approach sidesteps the absence of post-surgical own-voice audio and remains applicable to patients unable to phonate. Unlike existing alternatives (electrolarynx, esophageal speech, tracheoesophageal puncture), it can also restore the patient's own pre-surgical voice, offering a high-performing, low-complication path toward clinical use.

4.