Automatic System to Measure Speech Intelligibility (SAMI in French).
It is a blind unsupervised system designed to be close to the intelligibility of production and to the severity index of speech disorder measures.
Launching the Pase+_results.py file launch the experiments over the C2SI corpus.
All resulting experiments on this corpus can be found in results folder.
There you will find 3 subfolders (all, controls and files) corresponding to normalization using all features (all), using only controls features (controls) and using only the processed file (file).
All computations to the KL divergence are done in tools/pase_score.py module, the rest is in Pase+_results.py.
This work is under submission process.
The original parameters learned by Ravanelli et al. 2020 over 50h of the LibriSpeech dataset for the Pase+ model is under the trained_model/PASE+_parameters.ckpt file.
This repository depends on the audio_loader project and the PASE+ model.
The audio_loader and PASE+ projects have to be in the PYTHONPATH of the user.
All other requirements are listed in the pase+_environment.yml file.
This file can be used to create a pase+ environment with anaconda.
- Vincent Roger - [email protected]
- Jérôme Farinas - [email protected]
- Virginie Woisard - [email protected]
- Julien Pinquier - [email protected]