/jrm_ssl

Files for the paper: "Sound Source Localization using Deep Residual Learning"

Primary LanguagePythonMIT LicenseMIT

jrm_ssl

Files for the paper: "Sound Source Localization using Deep Residual Learning"

the programs run most on Python (Windows - Linux)

Requirements

Chainer - Install from pip

Hark - to obtain Audio Features

Training

Run ./chainer_train.py t -C $(config_file) from training folder to train a model

Evaluation

The forwarding file is located in the microcone folder

  • Run ./ssl_test.py $(DATE_OF_TRAINED_MODEL) to forward the audio files (Any corpus, any language is fine)
  • Run ./compile_results.py to obtain the block accuracy (median angle) - change the exp variable inside the file according to the folder you want to test
  • Run ./eval_correc_acc.py to obtain the point-to-point accuracy - change the exp variable inside the file according to the folder you want to test

Folder Structure

  • dataset_preparation : Two examples of the dataset prepared for the training
  • microcone : Files to evaluate any model and a network example to be trained
  • python_utils : extra files for training, preparing data, etc.
  • training : files for training a network
  • training_files : an example of a generated network and the files to test

Impulses Response

To generate the impulse use ISM of Eric A. Lehmann

Information of Microcone position microphones at HARK Supported Hardwares

Publication

JRM Vol.29 No.1 (Feb. 20, 2017)

License

This project is licensed under the MIT License - see the LICENSE.md file for details