- A state-of-the-art automatic speech recognition toolkit. Please refr to kaldi website for further information.
- refere to the kaldi-installation.md
Top-level directories : cmake COPYING egs misc scripts tools CMakeLists.txt docker INSTALL README.md src windows
- egs : examples, contains examples rrecipes that can be used as a template for future projects.
- src : source, contains most of the source code for programs that the training recipies call.
- Create a directory for your training data and your model
cd kaldi/egs
mkdir corpus
- Create the following directories:
mkdir exp conf data
mkdir data/train data/lang data/local
mkdir data/local/lang
will contain informartion about the audio files
- text: transcript of the corpus, following the formant: id_utterance WORD1 WORD2 ... WORDN
- Example of a text file: AA-test-F-S-AAS820006-00000 la formation professionnelle est un enjeu majeur tant du point de vue de nos performances économiques que du point de vue des idéaux de justice sociale
- segment: contains the start and end time for each utterance following the formant: id_utterance id_file start_time end_time
- Example of a segment file: AA-test-F-S-AAS820006-00000 AAS820006-test 0 8.6399375
- create a list of words in the corpus
cut -d ' ' -f 2- text | sed "s/' ' /\n/g" | sort -u > words.txt