ubclaunchpaddeploybot/minutes-prod repository overview

⁠minutes

Audio speaker diarization and transcription API.

⁠:running: Getting Started

⁠:rocket: Running the Server

make build-prod
make run

⁠:rainbow: Running the Research Environment

make build-dev
make dev

⁠:point_up: Deployment

We deploy continuously using Travis CI and a Docker Hub deploy bot.

You can also deploy the production and development images manually if you like.

make push-dev
make push-prod

You will need to set DOCKER_USERNAME and DOCKER_PASSWORD and be a member of the ubclaunchpad docker hub organization to deploy manually.

⁠:computer: Creating Training Data

We have a pipeline that is designed to take YouTube videos with transcripts and convert them into training data. pipeline.py is a CLI that will attempt to download the transcript and audio data for a given video, as well as prompt for some information that the rest of the pipeline uses to create labelled data (ie. what delimeters are used to identify speakers).

$ cd app/collector
$ ./pipeline.py <video_id>

⁠:point_right: Pushing Training Data

You can push training data into the research environment on DigitalOcean. You will need to collect the instance PEM file from your tech lead. Place the PEM locally in ~/.ssh/id_minutes. Set the environment variable MINUTES_RESEARCH_INSTANCE in your environment to the IP address of the DigitalOcean instance (available on Slack or from your tech lead).

Then, if you wish to push the file bigdata.csv, use the following command:

make FILE=bigdata.csv push-data

It will appear in the data folder on the research platform.