Goal:

1. Find a video on the OpenStreetMaps YouTube channel.
2. Download the audio of a talk.
3. Run it through OpenAI’s Whisper.
4. Send the transcript and the source URL to somebody in the OSM Community who has ownership over the OSM YouTube channel.

While you may be able to automate this, I don’t know how to do so.

What you need:

1. GPU (possibly NVIDIA, don’t know). 5gb vram (gpu ram). This might mean RTX 2060 or newer.
2. Strong cooling and noise isolation through building design.

Costs

1. Electricity will create some cost as transcription is hard. Do note that it is still less then the amount needed to power on and train a normal human being on the same task for several years in addition to the quantity of humans needed to get the same throughput.
2. This will result in wear and tare on your drives and other components.
3. This will make your computer and room warm in the summer. You need great cooling or the ability to use the excess heat for something valuable.

Steps:

1. Install Itch.io to assist updating.
2. Install whisper gui frontend by Grisk with Itch.
3. Download audio from a talk (not saying how).
4. Plug it in and get the result.
5. Send the URL of the talk and the transcript to unknownPerson who runs the OSM YouTube Channel in a standard format.

Sample format for an email

Hello noun, This email is to submit a transcript.

talk: https://www.youtube.com/watch?v=nsaiHhQvNSY model: whisper medium

Disclaimers:

1. I have yet to coordinate with anyone.
2. Human transcript writers are great and needed. They are in short supply. Let us reduce the net demand. They can save their energy for high impact legal and medical environments.
3. Maybe the built in YouTube transcript does the job well enough. This might not be worth the effort. I don’t know.