support M4A format as audio file

[AlirezaHaghi]

Description

Some providers, like Gemini, support audio formats such as M4A. link

But in Pydantic-ai, there is no way to upload such formats.

References

No response

[DouweM]

@AlirezaHaghi You can pass audio files to models including Google via https://ai.pydantic.dev/input/#audio-input. Is that not working for you?

[AlirezaHaghi]

From what I saw in the documentation, it appears they only accept a URL and don’t support local addresses.
And since in the end they inherit from the BinaryContent, I think they can’t have any new features.

But still, I tested it and I’m getting this error:

httpx.UnsupportedProtocol: Request URL is missing an 'http://' or 'https://' protocol.

Therefore, I don’t think AudioUrl is a suitable option.

[DouweM]

@AlirezaHaghi Ah, so the issue is with local URLs? The model wouldn't be able to use those URLs anyway right? If you'd like to use a BinaryContent, is the issue that audio/mp4 is not currently in messages.AudioMediaType?

[AlirezaHaghi]

@DouweM
For clarification and to prevent the XY problem: in my use case, the main problem is supporting M4A (not MP4).

It appears that the only way to pass local content into the AI models is through BinaryContent.

[DouweM]

@AlirezaHaghi Per https://stackoverflow.com/questions/39885749/is-a-m4a-file-considered-as-of-mime-type-audio-m4a-or-audio-mp4, .m4a should still use audio/mp4 as a MIME type.

It appears that the only way to pass local content into the AI models is through BinaryContent.

That's right, if the model supports file URLs, the <Foo>Url.url values are directly passed to the model, so it needs to be able to access the file.