Counting Filler Words and Detecting Upspeak on Your Smartphone (NES Group)

Great public speakers are made, not born. Practicing a presentation in front of a few colleagues (or a mirror) is common practice, and results in a set of subjective judgements what to improve. But can we estimate the quality of own delivery in a fair, unbiased, and repeatable way? We implemented Quantle - a smartphone-based presentation coach that constantly listens to the talk and estimates its effectiveness (see: https://bit.ly/2Th7sLT).

In this project we would like to extend this application by adding new features such as filler word counting (ahms, uhm, etc.) and upspeak detection (see: https://bit.ly/2EHCpQ7). We will help you to integrate your solutions into the existing code base. If time permits we may extend the app with even more features. Your ideas are very welcome!

Download as PDF

Student Target Groups:

  • Students of ICE/Telematics
  • Students of Computer Science

Thesis Type:

  • Master Thesis

Goal:

  • Gather data and create a machine learning model capable to detect simple filler words
  • Gather data and think about an algorithm how to detect upspeak;
  • Evaluate the performance of the proposed models on TED data set (provided);
  • If time permits, extend the coach with new features (bring your own ideas);
  • Present a demo and summarize the results in a written report.

Recomanded Prior Knowledge:

  • Creativity, interest in programming a mobile phone, speech processing, machine learning;
  • Good programming skills in Python, Java or C++;
  • Being familiar with speech processing algorithms is an advantage.

Start:

  • As soon as possible

Contact: