VoiceIdsShared omits fable, onyx, and nova documented for speech

Source report

Mirrors openai-java#642: generated speech voice constants omit fable, onyx, and nova.

Specification mismatch

The current CreateSpeechRequest.voice description lists fable, onyx, and nova as supported built-in speech voices. The referenced VoiceIdsShared enumeration omits all three. Generated Java SpeechCreateParams.Voice consequently has no corresponding constants.

The unrestricted string alternative still permits callers to supply these names manually, so this is an incomplete enumerated vocabulary and SDK discoverability problem, not rejection of all requests using these voices.

Expected correction

Make the speech voice vocabulary agree with the supported voices documented by CreateSpeechRequest, and regenerate affected SDK constants. If this shared vocabulary also serves APIs with a narrower set of voices, represent the endpoint-specific distinction explicitly.

Related report

#401 was closed after adding ash, coral, and sage. This report preserves the different missing voices identified in the Java issue. The source also questioned cedar and marin; the current speech description includes both, so that portion is already addressed.

Start in openapi.yaml at the CreateSpeechRequest.voice description and the VoiceIdsShared enumeration referenced in the issue. Compare the documented speech voices, update the vocabulary so fable, onyx, and nova are represented with any endpoint-specific distinction needed, then run the repository's generation workflow and verify SpeechCreateParams.Voice exposes the corrected constants.