Create speech-recognition-context.md

[yrw-google]

Add an explainer for the new speech recognition context feature

[padenot]

A very welcome addition, thanks for the initial design. Let's iterate a bit on the API shape, but I'm excited about this.

[padenot]

Is this just sequence<SpeechRecognitionPhrase> ?

[yrw-google]

I experimented with using sequence a bit. I think it is feasible to put sequence<SpeechRecognitionPhrase> inside SpeechRecognitionContext and get rid of SpeechRecognitionPhraseList, but that means we need to move other methods like the definitions of length (which is somehow required by blink/v8 if it detects an array) and addItem from SpeechRecognitionPhraseList into SpeechRecognitionContext too. In that case if we add more types of data to SpeechRecognitionContext in the future, it might become confusing, and we won't be able to support another array inside SpeechRecognitionContext because the definition of length needs to duplicate.

From the IDL examples I can find, seems like it is a common practice to create a new ObjectList interface to support a new Object interface (e.g. DataTransferItemList), and sequence is used more often in a dictionary. I'm not sure if we should make SpeechRecognitionPhrase and SpeechRecognitionContext become dictionary instead, so that their relationship will be as simple as SpeechRecognitionContext contains a sequence of SpeechRecognitionPhrase. We want to perform data validation on each SpeechRecognitionPhrase so using dictionary may oversimplify things? Let me know what you think!

[padenot]

What do we lose by simply doing:

Suggested change

	const recognition = new SpeechRecognition();
	recognition.start();
	var list = new SpeechRecognitionPhraseList();
	list.addItem(new SpeechRecognitionPhrase("updated text", 2.0));
	var context = new SpeechRecognitionContext(list);
	recognition.updateContext(context);
	const recognition = new SpeechRecognition();
	recognition.start();
	var list = [{text: "update text", weight: 2.0}];
	recognition.updateContext(list);

[padenot]

The same interface simplification goes for the other example.

[yrw-google]

If you still want to simplify things like this, can you tell me how IDL will look like in this case? We also want to perform data validation on each SpeechRecognitionPhrase, so using new SpeechRecognitionPhrase() helps us throw an error as we validate there, otherwise we would need to hold on validation until updateContext() is called?

[padenot]

Is this example leaving the actual use of the context that would lead to an error? I'm not sure if I follow.

[yrw-google]

I rewrote and added more sample codes to hopefully make it clear. Please take a look again and let me know if additional explanation is needed!