Kaldi setup for demonstration: in the end 1 audio file will be decoded. It uses docker and prebuild images, so you can run example, calling 1 shell script that will download images and run decoding; then you can learn images content and recreate example on your system without docker.
Setup of example is minimal: Kaldi + acoustic model. If you want something practical, check docker-kaldi-gstreamer-server. It's slightly more complex, but can be used to create real solutions, while this project is only for demonstration.
I hope it will be good starting point for someone who want to watch kaldi in action fast.
install docker
./run-and-decode.sh
It will download prebuilt images and then decode audio in audio-to-decode/ folder.
Images contain comments describing how to build and run kaldi. Images are huge, probably size can be reduced, pull requests are welcome.
images:
- kaldi: kaldi compilation from sources (~7GB)
- kaldi-model-en: based on kaldi image, contains english model and decoding script (+2GB over base image; overall ~9GB)
Go to kaldi tutorial and enjoy kaldi awesomeness