Content-Based Retrieval for Audio Files
Proposer: Audrey Mbogho
Brief Description: More and more universities are recording lectures. This is a very valuable method of preserving knowledge and widening its impact. Inevitably, as the volume of recorded material increases, searching for particular lectures will become a basic need. Similar to written documents, users will at times only recall some portion of the recorded content, rather than the date on which it was recorded or its title. Furthermore, a user may want to search for recordings that discuss certain topics, without knowing whether such recordings exist. This project will implement a method for retrieving recordings based on user-specified content.
Computer Science Content: Speech Processing, Pattern Recognition, Information Retrieval
Specific Learning Outcomes: Theoretical Analysis: Students will learn to apply hidden Markov models and/or neural networks to pattern recognition. Experimental Design: Students will learn to design experiments with sufficient breadth and depth to test a speech recognition system for robustness.
Skills Required by Team as a Whole:
Theory: Hidden Markov models, neural networks.
Implementation: Experimental design.
Facilities needed: Recording equipment, human speakers for training and testing.
Supervision: Audrey Mbogho
Number of Students: 2