Disfluency Removal API

Disfluency Detection, Removal and Correction: Increase Apparent Public Speaking Fluency By Speech Augmentation

• To run:

  • python audio_receiver_remover.py 7777
  • 7777 is the <port_number> to host the API
  • Open the index.html in Chrome/Firefox
  • Navigate to the tool & use the samples!

• If you change the <port_number>

  • Go to /js/vis.js and replace all http://localhost:7777/ with http://localhost:<new_port_number>/

• Required Packages:

  • web.py
  • librosa
  • tensorflow-gpu (>=1.5,<2)
  • pydub
  • pyAudioAnalysis
  • soundfile
  • h5py
  • scipy
  • numpy
  • sklearn
  • matplotlib
  • hmmlearn

Citation

If you use our tool please consider citing our paper:

@inproceedings{das2019increase,
  title={Increase Apparent Public Speaking Fluency by Speech Augmentation},
  author={Das, Sagnik and Gandhi, Nisha and Naik, Tejas and Shilkrot, Roy},
  booktitle={ICASSP 2019-2019 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)},
  pages={6890--6894},
  year={2019},
  organization={IEEE}
}

Contact

If you have any questions with the code please open an issue.