Carola Bonamico
From Biosignals to Words: Exploiting Novel Deep Learning Architectures for Speech Understanding.
Rel. Alessio Burrello, Daniele Jahier Pagliari, Giovanni Pollo. Politecnico di Torino, Corso di laurea magistrale in Ingegneria Informatica (Computer Engineering), 2026
|
Preview |
PDF (Tesi_di_laurea)
- Tesi
Licenza: Creative Commons Attribution Non-commercial No Derivatives. Download (6MB) | Preview |
Abstract
Silent speech (SS) interfaces decode biosignals of speech production into text without sound. One possibility is to use the neck surface electromyography (sEMG) that suits wearable devices and it is not invasive, capturing articulatory muscle activity even during silent mouthing. However, existing wearable systems based on sEMG recognize a few isolated commands, with limited evaluation after repositioning. This thesis extends them to closed-set classification and continuous recognition of sentences. The first contribution is a multimodal corpus including seven participants, six sessions collected at least one day apart and devices repositioned between them. Each participant recorded fifteen words and twenty sentences, ten times each, aloud and silently, for 29400 utterances over about 39 h 40 min.
We used two BioGAP-Ultra units recorded in parallel, one with a fourteen-channel sEMG neckband, one with a sixteen-channel EEG headband, both dry-electrode, each with a microphone to record synchronized EEG, sEMG, and audio signals
Relatori
Anno Accademico
Tipo di pubblicazione
Numero di pagine
Corso di laurea
Classe di laurea
Aziende collaboratrici
URI
![]() |
Modifica (riservato agli operatori) |
