A demo on automatic speech word selection and verification using Kaldi, with frontends.
- A basic POSIX environment(cygwin is ok), wget, zsh, sed and xz
- Kaldi : http://kaldi.sourceforge.net/
- Python, PySide and Matplotlib (only for PySide GUI front-end)
- Install all dependencies. Note that we need a full source tree of Kaldi, not only its binaries and libraries. You can find an example script of installing Kaldi under cygwin here.
- copy local.zsh.template to local.zsh, and point variables in local.zsh to the actual path in your environment.
- The PySide front-end is not tested under cygwin version of Python 3, you may need to install native Python under Windows (along with PySide, of course) and point the
PYTHON_EXECUTABLEvariable to the actual installation path of native Python. - Run 10.prepare/download_lexicon.zsh to get the CMU dictionary.
- The training of acoustic model used by this project involving Aurora-4 corpus and WSJ0 corpus, which are quite expensive if you just want to try it out. In this repository, there is already a trained model under model/, so you don't have to actually run the training script if you don't own the corpora.
- We directly use the Makefile from Kaldi. If there are bugs, things in this project will also be affected.
Yu-chen Kao (cybeliak) : cybeliak [the mysterious symbol] mail asteraceae org
This project is licensed under the Academic Free License, Version 3.0. See LICENSE for details.