MASTER THESIS ASSIGNMENT: NoisySSR - Silent Speech Recognition with Earables under Noise
INTRODUCTION
Noisy workshop-like environments hinder verbal communication. Earables equipped with IMU (Inertial Motion Units) and in-ear pressure sensors enable SSR (Silent Speech Recognition) for collaboration under noise.
OBJECTIVES
- Build and evaluate an SSR pipeline on earable data under controlled noise conditions;
- Compare classical ML baselines with CNN/LSTM/Transformer models;
- Quantify robustness across noise levels and task contexts with edge level deployment.
PROJECT DESCRIPTION
1. Literature Review: SSR for wearables/earables; noise-robust speech and articulatory sensing.
2. Data Collection: Data collection under various nose levels for workshop-level tasks
3. Modeling: A pipeline already exists from data collected in non-noisy environment. Improvement on this pipeline is expected.
4. Evaluation: Accuracy, F1, latency, robustness vs. SNR; user effort/usability notes.
5. Reporting: Reproducible code, experiment protocol, and deployment results.
PRE-REQUISITES
CS/Embedded/Data Science background; Python, PyTorch/TF, signal processing/ML basics.
WORK
20% Theory, 20% Experiments, 40% Programming, 20% Writing
CONTACT
Egemen İşgüder (egemen.isguder@utwente.nl)
Özlem Durmaz İncel (ozlem.durmaz@utwente.nl)
Rob H. Bemthuis (r.h.bemthuis@utwente.nl)