docker

Introduction

This Dockerfile builds a pre-configured Docker image for Estonian "general-purpose" speech recognition. In addition to speech-to-text, punctuation and speaker identification is also performed. Speaker identification models are built for persons who often appear in Estonian broadcast news.

Installation

The image is over 9 GB.

You can pull the image by running:

docker pull alumae/kaldi-offline-transcriber-et:latest

Usage

Start ta container (name is "speech2test") and put it into background (-d). Also, mount a local directory ~/tmp/speechfiles as the container directory /opt/speechfiles.

mkdir -p ~/tmp/speechfiles
docker run --name speech2text -v ~/tmp/speechfiles:/opt/speechfiles --rm -d -t alumae/kaldi-offline-transcriber-et

In order to transcribe a file, you have to place it to ~/tmp/speechfiles in your host machine and then invoke the /opt/kaldi-offline-transcriber/speech2text.sh inside the Docker container to transcribe the file. Note that the ~/tmp/speechfiles is equivalent to /opt/speechfiles from the container perspective.

Example:

cd ~/tmp/speechfiles
wget http://media.kuku.ee/intervjuu/intervjuu2018080910.mp3
# Note that you have to specify file paths from container's perspective
docker exec -it speech2text /opt/kaldi-offline-transcriber/speech2text.sh --trs /opt/speechfiles/intervjuu2018080910.trs /opt/speechfiles/intervjuu2018080910.mp3

The result (in Transcriber XML format) is now in ~/tmp/speechfiles/intervjuu2018080910.trs:

tail ~/tmp/speechfiles/intervjuu2018080910.trs
</Turn>
<Turn speaker="S1" startTime="268.11" endTime="298.79">
<Sync time="268.11"/>
eelmine aasta oli tegelikult selline lugu et jäi Balti matš ära Balti matš on toimub see aasta neljakümne teist korda selle aja jooksul on paar korda ära jäänud eelmine aasta oli Leedul üks niisugune asi mille pärast ta korraldada ei saanud ja eelmine aasta eelmine aasta ära ülemine aastal Eesti meeskondlikult
<Sync time="286.71"/>
kolmas aga väga tublid olid Eesti tüdrukud nii et et me just paistame sellega silma et et Eestis just nii-öelda tüdrukud ja naised on aktiivsed ja tegelevad selle tuletõrjespordiga
</Turn>
</Section>
</Episode>
</Trans>

Updating

If you want to replace the running container with a freshly pulled image, execute the following steps:

docker pull alumae/kaldi-offline-transcriber-et:latest
docker container stop speech2text
docker run --name speech2text -v ~/tmp/speechfiles:/opt/speechfiles --rm -d -t alumae/kaldi-offline-transcriber-et