fix(OpenAI): Handle diarized json response in TranscriptionResponseSegment

[olivier-drieux]

What:

• Bug Fix

Description:

TranscriptionResponse now parses both verbose JSON and diarized JSON responses.
Segment fields differ between the two response types.
This commit updates TranscriptionResponseSegment to handle both responses.
Note: Fields that differ between responses are now nullable, also, the id property can now be a integer or a string.

[iBotPeaches]

ah interesting. I have to read up on this - I missed it. Enabled CI to run for now

[iBotPeaches]

Thanks - looks good. I want to run it through some real testing when I have some time. I'll adapt one of my samples here and run it through on both ways to be sure - https://github.com/iBotPeaches/openai-php-laravel-test/tree/master/app/Console/Commands

[iBotPeaches]

Its odd the docs and your samples have non-null values, but my raw tests return a lot more nulls than I expect. Below is a real call sample

{
	"text": "The quick brown fox jumps over the lazy dog.",
	"segments": [
		{
			"type": "transcript.text.segment",
			"text": " The quick brown fox jumps over the lazy dog.",
			"speaker": "A",
			"start": 0,
			"end": 2.1499999999999995,
			"id": "seg_0"
		}
	],
	"usage": {
		"type": "tokens",
		"total_tokens": 172,
		"input_tokens": 51,
		"input_token_details": {
			"text_tokens": 11,
			"audio_tokens": 40
		},
		"output_tokens": 121
	}
}

[iBotPeaches]

Tested it out here - iBotPeaches/openai-php-laravel-test@acc7a03

Confirmed failure prior to patch and success afterwards. I don't fully understand why I have nulls on a ton of fields, but direct calls via CURL replicate as well.

[iBotPeaches]

thanks!