Imaginary Soundscape Demo

Overview

• This is a demo repository of "Imaginary Soundscape" project by Qosmo, Inc..
• On the project website, you can add AI-generated soundscape to any images you upload or specify. You can try the same process with this repository.
• Our model is based on SoundNet, proposed in 2016.
• Our code is written in Python3 using keras and librosa.

How to use

1. Please download this zipped data file and unzip it into the same directory as playground.ipynb.
2. Open playground.ipynb with Jupyter notebook.

References

• Imaginary Soundscape website
• Original demo of Imaginary Soundscape
• Short paper at NIPS2017 Workshop
  "Imaginary Soundscape: Cross-Modal Approach to Generate Pseudo Sound Environments"
  Yuma Kajihara, Shoya Dozono, Nao Tokui
• Imaginary Soundscape — Take a walk in soundscapes “imagined” by AI - A post on medium.com

Model

Prerequisites

• Python 3 (>= 3.6)
• Jupyter Notebook
• keras (2.x)
• librosa
• scipy
• Linux or MacOSX

Credits

• Concept/Machine Learning: Nao Tokui @naotokui
• Machine Learning: Yuma Kajihara @kajiyu
• UI Design/Programming: Shoya Dozono and Robin Jungers @robinjungers