Computes Short Term Objective Intelligibility measure


License
MIT
Install
pip install pystoi==0.4.1

Documentation

Python implementation of STOI

Implementation of the classical and extended Short Term Objective Intelligibility measures

Intelligibility measure which is highly correlated with the intelligibility of degraded speech signals, e.g., due to additive noise, single/multi-channel noise reduction, binary masking and vocoded speech as in CI simulations. The STOI-measure is intrusive, i.e., a function of the clean and degraded speech signals. STOI may be a good alternative to the speech intelligibility index (SII) or the speech transmission index (STI), when you are interested in the effect of nonlinear processing to noisy speech, e.g., noise reduction, binary masking algorithms, on speech intelligibility.
Description taken from Cees Taal's website

Install

pip install pystoi or pip3 install pystoi

Usage

import soundfile as sf
from pystoi import stoi

clean, fs = sf.read('path/to/clean/audio')
denoised, fs = sf.read('path/to/denoised/audio')

# Clean and den should have the same length, and be 1D
d = stoi(clean, denoised, fs, extended=False)

Running the Octave tests

sudo apt update 
sudo apt install octave octave-signal 
pip install oct2py
python -m pytest tests/test_python_octave.py
python -m pytest tests/test_stoi_octave.py

Matlab code & Testing

All the Matlab code in this repo is taken from or adapted from the code available here (STOI – Short-Time Objective Intelligibility Measure – ) written by Cees Taal.

Thanks to Cees Taal who open-sourced his Matlab implementation and enabled thorough testing of this python code.

If you want to run the tests, you will need Matlab, matlab.engine (install instructions here) and matlab_wrapper (install with pip install matlab_wrapper). The tests can only be ran under Python 2.7 as matlab.engine and matlab_wrapper are only compatible with Python2.7 Tests are passing at relative and absolute tolerance of 1e-3, which is enough for the considered application (all the variability is coming from the resampling method when signals are not natively sampled at 10kHz).

Very big thanks to @gauss256 who translated all the matlab scripts to Octave, and wrote all the tests for it!

Contribute

Any contribution are welcome~, specially to improve the execution speed of the code~ (thank you Przemek Pobrotyn for a 4x speed-up!) :

  • Improve the resampling method to match Matlab's resampling in tests/. This can be considered a solved issue thanks to @gauss256 !
  • Write tests for Python 3 (with transplant for example)

References

  • [1] C.H.Taal, R.C.Hendriks, R.Heusdens, J.Jensen 'A Short-Time Objective Intelligibility Measure for Time-Frequency Weighted Noisy Speech', ICASSP 2010, Texas, Dallas.
  • [2] C.H.Taal, R.C.Hendriks, R.Heusdens, J.Jensen 'An Algorithm for Intelligibility Prediction of Time-Frequency Weighted Noisy Speech', IEEE Transactions on Audio, Speech, and Language Processing, 2011.
  • [3] J. Jensen and C. H. Taal, 'An Algorithm for Predicting the Intelligibility of Speech Masked by Modulated Noise Maskers', IEEE Transactions on Audio, Speech and Language Processing, 2016.