Clarify input size is implementation-defined

[dotun-code-god]

What is the limit to the content that can be passed to the audio input and what is the maximum duration?

Also, can anyone recommend a chunking mechanism to use to optimize larger audio files?

[tomayac]

(This is an implementation-specific detail, not a spec issue, which this repo is for, so please refrain from posting such issues in the future. Instead, join the early preview program where questions like this are welcomed on the discussion list.)

In Chrome's implementation, the current limit is 30s of audio. See the article for links to examples. Here's a simple working demo (courtesy of an LLM) that shows splitting audio files in 30s chunks.

[anssiko]

That's a good question, this spec repo is just a wrong place for it. @tomayac has good pointers.

For the spec, we could improve by stating what is https://infra.spec.whatwg.org/#implementation-defined

For this limit, we could update the note at the top of https://webmachinelearning.github.io/prompt-api/#prompt-processing or put it inline in the algorithm. A task for a future editor :)

[dotun-code-god]

Yeah, that's very true. I'm sorry for not putting that into consideration.

However, thank you for your guidance. @tomayac and @anssiko