Add Explainer for On-Device Speech Recognition

[evanbliu]

This PR adds an explainer for on-device speech recognition support.

[yashrajbharti]

Hi, this is an exciting approach! I've added a point on sustainable web highlighting the environmental benefits of on-device processing.

I got to know from Kenji about the focus on supporting MediaTrack and on-device ASR. I wanted to confirm if this explainer also aims to cover those aspects, as I believe I could contribute here. If not, I wouldn't want to dilute the main message.

[evanbliu]

Thanks for the suggestion @yashrajbharti! In this case, on-device speech recognition might not always have a reduced environmental impact. If a user rarely uses the Web Speech API, the bandwidth savings of on-device speech recognition might not outweigh the cost of downloading the on-device speech recognition models. Also, there it isn't exactly an apples-to-apples comparison between cloud and on-device speech recognition in terms of energy consumption and there are many factors that affect this. In Chrome for example, the cloud service that powers speech recognition is a bit antiquated but runs in highly optimized data centers whereas the local on-device speech recognition runs newer cutting edge models but has varying power consumption levels depending on the user's device. Because this isn't such a clear cut advantage, I suggest omitting it as an explicit benefit of using on-device speech recognition.

[yashrajbharti]

Thanks for the detailed insight @evanbliu! While reducing network requests generally contributes to sustainability, I understand that the impact can vary. Without data to support average estimates, drawing a clear-cut advantage would indeed be difficult.

[evanbliu]

@hoch - Can you please take a look at this PR when you get a chance? Thanks!

[hoch]

LGTM

Perhaps we can restructure the directory/file?

explainers/on-device-speech-recognition.md

Or are you planning to have more supporting materials for the on-device feature?