1

I am working on an application which uses Vosk for speech recognition. I would like to create a dictionary for the application which contains only the trigger words and spoken numbers needed by the application. Using command line instructions found here: www.alphacephei.com/vosk/adaptation I was able to install Kaldi on my laptop. These are,

export KALDI_ROOT=`pwd`/kaldi
git clone https://github.com/kaldi-asr/kaldi
cd kaldi/tools
make
extras/install_opengrm.sh

However, I am having a problem building a dictionary using the provided commands. These are,

export PATH=$KALDI_ROOT/tools/openfst/bin:$PATH
export LD_LIBRARY_PATH=$KALDI_ROOT/tools/openfst/lib/fst
cd model
fstsymbols --save_osymbols=words.txt Gr.fst > /dev/null
farcompilestrings --fst_type=compact --symbols=words.txt --keep_symbols text.txt | \
ngramcount | ngrammake | \
fstconvert --fst_type=ngram > Gr.new.fst
mv Gr.new.fst Gr.fst

The problem occurs at "cd model" because there is no /model directory in the directory structure created during the Kaldi installation. Checking in my Vosk project, I find /models, but no /model directory either.

I have tried creating /model in /kaldi/tools and then running the above commands with no success. Please let me know what I am missing here. Thanks in advance.

portsample
  • 1,986
  • 4
  • 19
  • 35

1 Answers1

1

The command cd model in the docs is actually incomplete. To run this you have to cd into the directory where Gr.fst exists. This file usually exists in the directory <any model with dynamic graph>/graph.

  1. Head to https://alphacephei.com/vosk/models, download a model that supports dynamic vocabulary reconfiguration (usually small models do, big models don't).
  2. Unzip the folder
  3. Prepare a .txt file with words that your project relates to.
  4. Proceed with the steps mentioned in your second code snippet (with a slight modification to the cd model part)
Scotrraaj Gopal
  • 81
  • 1
  • 6