Skip to content

Latest commit

 

History

History
30 lines (24 loc) · 1.19 KB

README.md

File metadata and controls

30 lines (24 loc) · 1.19 KB
Raw

kaldini

Train and test an ASR from scratch using the kind of data that you can get from a small group of native informants in a new language in about five hours of contact time, i.e., only about 1000 read sentences.

Usage:

git clone https://www.github.com/kaldi-asr/kaldi
cd kaldi/tools; make -j $(nproc)
cd ../src;      ./configure --shared && make depend -j $(nproc) && make -j $(nproc)
cd ..;          mkdir egs
git clone https://www.github.com/uiuc-sst/kaldini
cd kaldini

The goals of this project:

  1. Train and test an ASR from scratch using the kind of data that you can get from a small group of native informants in a new language in about five hours of contact time, i.e., only about 1000 read sentences.

  2. Explore the use of python to train and test kaldi recognizers, bypassing some of Dan Povey's heavily tested but obscure bash scripts. Sometimes my python scripts call his bash scripts, just because often that's the most efficient way to make it work. This is all really experimental.

Anyway, the python code is in the python subdirectory. The script run.sh still points to bash versions of the same thing. I don't remember which ones run and which ones don't, right now.