Project Abstract

Laughter detection is for the most part an unexplored area of audio classification, giving systems a tool to detect laughter in clips of audio can facilitate more nuanced data collection and how a system can interact with it’s users. The main goal of this project is to create a machine learning model that can predict whether or not an audio file has laughter present. The project was worked on using python with various mahcine learning and audio processing packages (tensorflow, sklearn, librosa, etc). Various audio features were taken from a labelled dataset made up of over 100 hours of audio. The features and labels were fed into a machine learning model which resulted in a model that can predict laughter.The main result of this project is that it is possible to find the difference between laughter and speech by extracting certain audio features (including MFFCs, RMS, zero-crossing-rate, ), however the results of this project show that a large labelled dataset with a balanced ratio of laughter and speech is required for more reliable results. In conclusion the model created from this project can detect laughter to a degree, however more training data with a better balance between laughter and speech would be needed to further increase the accuracy.

Keywords: Machine Learning, Deep Neural Networks, Audio Processing

 Conference Details

Session: Poster Session A at Poster Stand 100

Location: Sir Stanley Clarke Auditorium at Tuesday 7th 13:30 – 17:00

Markers: Julian Hough, C�cilia Pradic

Course: BSc Computer Science, 3rd Year

Future Plans: I’m looking for work