No items found.
Pops Presentation Advisor
Machine learning service evaluates video presentations, extracting key features, offering feedback to speakers, used by thousands of students and professionals.
Client
Advanced Solutions
Date
July 1, 2020
Service
Machine Learning Engineering
Location
Louisville - KY (Remote)
Challenge
The project involves developing a machine learning-powered service for evaluating public speaking presentations in video format. The challenges include:
- Complex Video Preprocessing: Extracting audio, video, and speech to
text data from raw video requires intricate processing techniques. - Data Analysis and Feature Engineering: Utilizing tools like Pandas, Matplotlib, and advanced models to analyze data and engineer relevant features.
- Custom Deep Neural Network Training: Training a deep neural network using GPU acceleration with a limited dataset demands careful optimization and tuning.
- Body Pose Extraction: Extracting presenter body pose for meta-features and suggestions adds complexity to the feature extraction process.
- Deployment and Integration: Deploying the machine learning model and creating a functional HTTP API on the Innovation Lab Fabric platform necessitates seamless integration and testing.
Project Results
The efforts resulted in the creation of an efficient machine learning-powered service. Through complex video preprocessing, data analysis, and feature engineering, the service accurately evaluates public speaking presentations. The custom deep neural network, trained with optimized GPU processing, ensures precise analysis despite limited data. Presenter body pose extraction enhances feedback quality, leading to improved presentations. The seamless deployment and integration on the Innovation Lab Fabric platform have enabled thousands of students and professionals to receive valuable feedback, enhancing their public speaking skills significantly.