How Might We Help Teachers to Practice Presenting?
EduTalk
Enhancing presentation skills can be a difficult task for teachers, especially for student teachers. To tackle this challenge, this project integrated AI into website design. The goal of this project is to provide helpful suggestions to student teachers while also making AI more transparent and understandable.
AI in Design
Explainable AI
Platform Design
TIME
09/ 2022 - 11/2022 (3 months)
LOCATION
Eindhoven University of Technology
TEAM
J.R. van den Broek, T.T.H. Poell, Y.Y. Lo, J.M. Zandbergen
SKILLS
Field Research, Stakeholder management, Workshop facilitation, Service design
Background
Among different types of lectures, it is generally accepted that students' attention declines during a long lecture. Teachers' presentation skills can be crucial to keep students focused on the lecture's material.
According to the literature, we found that student attention is best piqued and maintained by appropriate vocal inflection and expressiveness.
Problem Statement
Improving Presentation Skills is Challenging for Teachers.
Insufficient Feedback for Practice
When the student teacher graduates, the new teacher does not have a supervisor providing feedback. Starting as a teacher is described as highly chaotic and demanding.
Seldom reflect on previous courses
In practice, we see that reflection is little done. Teachers prefer to spend time preparing new lectures instead of looking back and reflecting on already given lectures.
Ideation Process
Design Concept
To make reflecting after lectures easier for starting teachers and gradually achieve a lively voice, we proposed EduTalk, an online system. It visualizes the change in speaking speed and intonation and also provides suggestions for teachers.
Record the lecture
Receive suggestions after class
Design highlights
01
Design with AI to Give Feedback to Teachers
Utilizing AI can help teachers to practice presenting by themselves. Teachers can receive suggestions online and see their progression.
02
Make data explainable and transparency
Teachers can review visualization of intonation and speaking speed of their previous presenting and understand reasons for suggestions.
Website Design
Interaction Flow
Design Evaluation
Validation of AI model
The data set only had 30 data entries, the decision was made to focus primarily on training.
From Graph be low, the optimal value for K seems to be within 6-10. However, due to small data entries, this already seems to approach overfitting. More data is needed to be able to find the optimal value for K.
Graph for the cross-validation
Not Well-Distributed Data Set
The major issue is that some values were never even labelled correctly.
Confusion matrix from testing
Overall Accuracy 25%
From the the output from the confusion matrix which shows that overall this performance is abysmal. But it all comes down to not having enough data points in the model.
Output from the confusion matrix
Discussion
Low accuracy of the supervised learning algorithm
To improve accuracy, more audio files of lectures should be analyzed and fed into the algorithm in further work.
Various types of lectures could be considered.
Current Edutalk would over-generalizing various lectures, additional labels could be added to the data. For example, the desired average speaking speed of an information-packed lecture may be different from an open-discussion lecture.
Reflection
While discussing the user flow, we reviewed the literature to find previous research papers that support our concept. Although the concept of “liveliness” and its relation with speech rate, speed changes, and annotations should have been better defined and supported with literature. I learned to timely throw back our design contexts and iterated design brief instead of continuing the prototype and making a sophisticated interface.
© YIN-YU LO ALL RIGHTS RESERVED