This is a project to teach a machine to provide canned laugh tracks to normal, everyday life. The goal is to use Machine Learning to listen to an audio stream and identify when the optimal time is to insert some laughter.
The project is separated into multiple phases, aimed at building increasing levels of understanding.
First, we need to generate a corpus of laughter.
VGGish is a pretrained model for identifying audio.
There is a Dockerfile to make it easy to spin up. Build and run with:
docker build -t laugh-tracks .
nvidia-docker run -it --rm -v $(pwd):/notebooks/ -p 8889:8888 --name laugh-tracks laugh-tracks
or in fish shell:
docker build -t laugh-tracks .
4FA2
nvidia-docker run -it --rm -v (pwd):/notebooks/ -p 8889:8888 --name laugh-tracks laugh-tracks