Kaldi

Kaldi-Docker

This is a docker wrapper around Kaldi.

Docker

docker run -v <local data directory>:/data/ -p 8888:8888 springernlp:chapter_8k:latest /bin/bash

If a GPU is available, run:

docker run -it --runtime=nvidia -v /data/:/data/ -p 8888:8888 springernlp/chapter8k:gpu /bin/bash

Start the python notebook.

jupyter notebook --no-browser --ip=0.0.0.0 --port=8888 --allow-root --NotebookApp.token=''

Go to localhost:8888.

Build the Docker Image from Scratch

Additional requirements due to licenses:

SRILM: SRILM has an more restrictive license and therefore requires you to fill out a form. Download SRILM from here. change the name of the file to srilm.tgz and move it into the root directory of this repository.

/kaldi-docker/srilm.tgz

The Dockerfile will add this to the image build.

Warning: If you are building this image from scratch, it can take a very long time (to the tune of an hour).

docker build -t chapter_8k:latest .

Running the Common Voice example

• Source /kaldi/tools/env.sh
• Change data directory in /kaldi/egs/commonvoice/s5/run.sh
• If dataset is downloaded already, make sure there is a .complete file in the cv_corpus_v1 directory (touch .complete). Otherwise it will download the dataset again.
• Change cmd.sh from queue.pl to run.pl (don't use distributed queue library)

cd /kaldi/egs/commonvoice/s5/
./run.sh

TODO: [x] Change to correct directory [x] Must change run.sh for dnn run (it sources that file before running) [] Still haveing to run the install at the beginning of the file for now.. not sure why [] Copy notebook into dir [] Split cpu and gpu containers appropriately [] Push GPU container jupyter notebook --ip=0.0.0.0 --allow-root