With the emergence of powerful speaker recognition systems like Alexa and Siri, people are getting used to give voice commands instead of resorting to other physical inputs (e.g., keyboards, buttons). Together with the rapid growth of the Internet of Things (IoT), an increasing number of embedded devices are being deployed in various settings, and these can highly benet from the ability of listening to their surrounding and gain context awareness. However, low-power IoT devices are often not as powerful as Alexa or Siri, only have a fraction of their memory and processing power, and cannot always rely on a cloud service to perform heavy computations (e.g., machine learning tasks).
Our goal is to let low-power IoT devices (i.e., deeply embedded devices with limited memory and computational resources) perform speaker distress detection autonomously: this could be used to automatically look for help when distress is detected in a person's voice. Within our group, we have built a low-power custom earpiece embedding cheap microphones and others sensors that can be used to carry out this task. Our aim is to study how state-of-the-art systems based on machine learning techniques can be shrunk to t the constraints of low-power IoT devices, and evaluate whether the shrunk models can still sustain a high detection accuracy.
Download as PDF
Student Target Groups:
- Students of ICE/Telematics;
- Students of Computer Science;
- Students of Electrical Engineering.
Thesis Type:
- Master Project / Master Thesis / Paid Student Job
Goal and Tasks:
- Understand how state-of-the-art distress detection systems work, and investigate how they can be shrunk to t on tiny embedded devices;
- Develop a novel lightweight distress detection model/scheme that ts even on the smallest IoT devices;
- Develop a prototype of a distress detection system running on a constrained IoT device (e.g., our custom earpiece embedding the nRF5340).
- Leverage sensors such as heart rate and/or pulse oximeter sensors to create very accurate and personal distress detection devices.
Recommended Prior Knowledge:
- Basic knowledge of machine learning;
- Solid skills in Python and C programming;
- Experience with microcontrollers.
Start:
Contact: