Skip to content

Latest commit

 

History

History
13 lines (10 loc) · 745 Bytes

README.md

File metadata and controls

13 lines (10 loc) · 745 Bytes
Raw

Automatic Speech Intelligibility Prediction From Single Channel Speech Signals

This repository contains the code for:

  • dataset generation: add noise to clean signals LibriSpeech dataset;
  • dataset generation: compute reliability/intelligibility index by comparing automatic transcription and original text;
  • feature extraction: compute a set of audio features (MFCC, ZCR, SC, ...)
  • model training: train supervised classifiers for intelligibility prediction

References

"Automatic Reliability Estimation for Speech Audio Surveillance Recordings" Clara Borrelli, Paolo Bestagini, Fabio Antonacci, Augusto Sarti, Stefano Tubaro 2019 IEEE International Workshop on Information Forensics and Security (WIFS)