Why a Silent-Speech sEMG Recording Session Fails: Practical Failure Modes of a Low-Cost Wearable Prototype
Work on silent speech based on surface electromyography (sEMG) reports recognition accuracy but rarely reports what actually goes wrong during recording on low-cost hardware. This paper collects the failure modes measured across six recording sessions of a single wearable prototype between September and October 2026. Five classes of failure are described: mains interference from a laptop charger, dry electrode contact, insufficient articulation effort, speaker fatigue, and saturation of unconnected inputs. Each is given with a measured magnitude and with a check that can be run before recording begins. Mains interference from a charger plugged into the wall occupied 97–99% of signal power in the working band; switching to battery power reduced its share to 2%. Dry electrodes produced a resting noise of about 53 µV, identical across seven channels, which points to a common interference source rather than physiology; after wetting, noise fell to 2–5 µV. Silent articulation at ordinary effort did not differ from silence in amplitude (7.50 against 7.60 µV), whereas exaggerated articulation did. In two of six continuous recordings, between two and five of eight channels were saturated for the entire session. A pre-recording check sequence is proposed that rules out each failure in minutes and requires no equipment beyond the prototype itself. Russian and English versions attached.
Authors
- Fedotova Ann (ORCID: https://orcid.org/0009-0008-7300-3460)
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-16
- DOI
- https://doi.org/10.5281/zenodo.24279763
- Primary Topic
- Muscle activation and electromyography studies
- Type
- preprint