https://hal.archives-ouvertes.fr/hal-02963802
Raw File
Tip revision: 6b9bf3964385d0c16d262796d9e4a3a30a52dafd authored by Software Heritage on 12 October 2020, 00:00:00 UTC
hal: Deposit 1045 in collection hal
Tip revision: 6b9bf39
README
In order to run the speech recognition baseline, please:

- Download and extract the voiceHome-2 corpus (https://doi.org/10.5281/zenodo.1252143)
- Download and install the FASST toolbox (https://gitlab.inria.fr/bass-db/fasst)
- Download and install Kaldi (https://github.com/kaldi-asr/kaldi)
- Download and extract the acoustic model (https://doi.org/10.5281/zenodo.4079314)
- Run main_compute_est_loc.m to estimate DOAs (see localization baseline)
- Run main_compute_speaker_models.m and main_separation.m to enhance speech (see enhancement baseline)

Then:

1) Rename the AM folder as voiceHome_scripts_dict/expReverb/small_set_rev_with_noise/tri4a_dnn_tr/.
2) Run decode_interface.sh.

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% Copyright 2017 Irina Illina, Emmanuel Vincent (for domotic_multimedia_grammar_10aout_emmanuel_corrected_ver4.jsgf)
% Copyright 2017 Sunit Sivasankaran (for all other files)
%
% This software is distributed under the terms of the Apache License
% Version 2.0
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
back to top