A Deep Reinforcement Learning Approach to Audio-based Navigation in a Multi-speaker Environment (ICASSP 2021)

This is the repository for our paper A Deep Reinforcement Learning Approach to Audio-based Navigation in a Multi-speaker Environment accepted in ICASSP 2021.

You can download the Unity project package here.

The downloaded AudioRL.tar.xz contains the Unity project, which includes the Unity environment, the Agent code and trained model, and the dataset. AudioRL contains the agent configuration file (ListeningAgent.yaml).

Installation

Follow the Unity ml-agents installation instructions here.

Running the environment

1. Open the AudioRL Unity project
2. Click Play on Unity Editor window to run a pretrained Agent in the environment.