I'm writing a TAG comment on an extension to the Web Speech API. One user of the API is captioning speech in a video conference system. My understanding is that they currently intend to

• have each recipient run speech-to-text on the audio they receive.

Two other possible architectures are that

• each sender runs speech-to-text on the audio they're sending, so the text can be shared between all recipients, or
• the server runs speech-to-text centrally in their data center.

I was hoping to find a principle in the Sustainability Guidelines that would cover this, which I could cite in my comment, but I don't see one. For example, you could say to "avoid redundant processing", and you could discuss the rough difference in energy efficiency between computing something on a server vs a client.