When using sdk/openai, important data is discarded by the SDK

[adolphnov]

Description

When I use the gpt-4o-audio-preview model and pass the audio, the fetch method I've implemented myself, chat with openai to require the audio and transcription, the audio in the returned data was discarded, rendering the entire conversation completely invalid.

Code example

No response

AI provider

@ai-sdk/openai v1.0.4

Additional context

I found that the zodSchema.safeParse method dropped the audio field.

[lgrammel]

Can you provide details on which information you would need to be returned? currently only audio inputs are supported

[adolphnov]

Can you provide details on which information you would need to be returned? currently only audio inputs are supported

I have input the audio and passed in additional parameters myself:

readonly fetch = async (url: RequestInfo | URL, options?: RequestInit): Promise<Response> => {
        const body = JSON.parse(options?.body as string);
        if (body.model === 'gpt-4o-audio-preview') {
            body.modalities = ['text', 'audio'];
            body.audio = { voice: 'alloy', format: 'opus' };
        }
        return fetch(url, {
            ...options,
            body: JSON.stringify(body),
        });

The original response from OpenAI contains an audio field:

{
  "index": 0,
  "message": {
    "role": "assistant",
    "content": null,
    "refusal": null,
    "audio": {
      "id": "audio_abc123",
      "expires_at": 1729018505,
      "data": "<bytes omitted>",
      "transcript": "Yes, golden retrievers are known to be ..."
    }
  },
  "finish_reason": "stop"
}

However, it was filtered by the SDK internally and the audio field was discarded.

I need the original return result, not the trimmed data. The response data I get in the onStepFinish middleware is also filtered and doesn't include the audio field.

[polesapart]

Ideally, ai sdk should formally support audio, as it supports images and pdf. Other ai providers may as well implement audio i/o soon enough, i.e. anthropic & hume ai colaboration rumours are around.