This module is produced as part of the requirement for my Final Year Project for my BSc Computer Science degree requirement at the University of Portsmouth. This module represents the second round of experimentation for my research project which involves producing a speech recognition system for sEMG silent speech.
The inspiration for this project largely comes from the landmark sEMG silent speech synthesis papers "Digital Voicing of Silent Speech" at EMNLP 2020 and "An Improved Model for Voicing Silent Speech" at ACL 2021. These papers provided a method to transduce silent speech surface EMG (sEMG) signals directly into speech features (either MFCCs or mel spectrograms). This project goes one step further and uses the predicted mel spectrograms from the transduction model and directly performs speech recognition on these speech features to produce a highly efficient silent speech recognition system.
The dataset for this project is based on the open-source dataset released with the "Digital Voicing of Silent Speech" at EMNLP 2020 paper with the additional force-aligned phonemes released along with the "An Improved Model for Voicing Silent Speech" at ACL 2021 paper.
The EMG and audio data can be downloaded from https://doi.org/10.5281/zenodo.4064408.
And the force-aligned phonemes for the dataset (not including the closed vocabulary portion of the dataset) can be downloaded from https://github.com/dgaddy/silent_speech_alignments/raw/main/text_alignments.tar.gz.
You can clone this repository by doing the following:
git clone https://github.com/MiscellaneousStuff/semg_asr.git
git submodule init
git submodule update