Hive-Transcribe
A PeerTube plugin generating subtitles for your videos via Automatic Speech Recognition.
Powered by Vosk.
Forked from PeerTube Transcription Plugin.
Installation
After plugin install, Peertube needs to be restarted!
[DEV MODE] Install deps and build
yarn
yarn build
[DEV MODE] Install deps, build and install on local peertube
Copy buildAndDeploy.template.sh to buildAndDeploy.sh and customize peertube credentials
yarn
./buildAndDeploy.shPeerTube UI
Browse to Administration > Plugin/Themes > Search and type hive in the search bar:
Click Install then Confirm.
Yarn (or npm) script
yarn plugin:install -n peertube-plugin-transcriptionIn production you may have to specify env variables:
cd /var/www/peertube/peertube-latest/ NODE_ENV=production NODE_CONFIG_DIR=/var/www/peertube/config yarn plugin:install -n peertube-plugin-transcription
PeerTube CLI
Setup PeerTube CLI: https://docs.joinpeertube.org/maintain-tools?id=installation
peertube plugins install --npm-name peertube-plugin-transcriptionUsage
The video transcription happens when you Update a video and set its language.
The process is executed in the background and, in our tests, lasted for half the time of the video.
For example, it took approximately 5 minutes to transcript a 10 minutes long video.
There are currently no notifications to warn the user when the process is over (see Chocobozzz/PeerTube#5076. If it succeeded, a caption should be present in the Captions tab of the video edition page.
If not, feel free to fill an issue.
By default, a light language training model is used, so you may be disappointed by the quality of the transcript and may wish to choose a larger model for better results (see below).
Settings
From the transcription plugin settings page, you may choose a language training model for the available languages.
You'll probably need to configure only the few you'll be using on your instance since some of them may weight up to 2 Gb!
Only languages with available training are currently proposed. Checkout the following page for a better description of each available models: https://alphacephei.com/vosk/models
Documentation
About
Initial codebase developed with the financial support of the Direction du Numérique pour l'Éducation (DNE) for the ministry of Éducation nationale, de la Jeunesse et des Sports (french ministry of education) in the context of the apps.education.fr initiative.