Student: Inge Salomons
Supervisors: Eva Navas Cordón e Inma Hernáez Rioja
DESCRIPTION
In recent years, Silent Speech Interfaces (SSIs) have emerged as a
promising alternative to restore oral communication by decoding
speech from non-acoustic (silent) speech-related biosignals
generated during speech production. Electromyography (EMG),
which captures facial muscle activity using surface electrodes,
offers a fundamentally new solution to restore communication
capabilities to speech-disabled persons. In this approach, audible
speech is directly generated from silent speech data by mapping
the EMG generated signals into a suitable speech representation
and then generating a waveform from the estimated speech
parameters. Most commonly, deep neural networks (DNNs) are
applied to model the silent speech-to-speech mapping.
This thesis will be developed in the framework of the “Voice
Restauration with Silent Speech Interfaces (ReSSint)” project,
funded by the Agencia Estatal de Investigación in collaboration
with the University of Granada and the Cognitive System Lab of
the University of Bremen.