chore: add missing openai.fm model names to TTSModelSettings.voice type

[hironow]

changes

Include two missing model names ( ballad , verse ) from openai.fm in the voice type definition to ensure coverage of all TTS voices.

Note that a simultaneous update of the dependency module is required for full compatibility—this is the reason why tests are failing.
https://github.com/openai/openai-python/blob/6d0ecdd8ecbface903cf93c7571398b90b803b0b/src/openai/resources/audio/speech.py#L56

[rm-openai]

looks like tests are failing?

[hironow]

Thanks for your review! @rm-openai

I realized that I omitted some details in the description.
It appears that the test failures are due to this process, as explained here (occurred type mismatching):

To clarify, the changes work together with the related PR in openai-python(deps repo) as follows:

1. The model type is updated in openai-python (deps repo and chore: add missing openai.fm models to voice Literal openai-python#2237 ).
2. The updated version is then used as a dependency in openai-agents-python (this repo).
3. Finally, the model type is updated in openai-agents-python (this repo and this PR).

Please note that first, the addition of the model type on the openai-python side must be completed. After that, when openai-agents-python uses the updated version (step 2), there is a concern that the typecheck might still fail.

I appreciate your attention to this matter!

[rm-openai]

Makes sense. I think we should probably update this param in the agents SDK to be a str | Literal, so that you can pass arbitrary model names

[hironow]

That makes sense. However, even with the parameter updated to str | Literal, passing a string now causes an error. Should we wait for an update on the openai-python side, or use a type: ignore workaround for now?

suggestion here:

voice=settings.voice or DEFAULT_VOICE, # type: ignore

[rm-openai]

yes, we'll be updating the openai SDK soon