Why does Safari on iPhone send an unsupported audio file format to Whisper API?

Below is a likely explanation and some recommendations:

Explanation

Although you’re explicitly setting the MIME type to audio/wav in your front‑end, Safari on iOS may be encoding the recording in a slightly different container/format (for example, generating an m4a file) or even missetting the actual file extension when building the blob. When you save the temporary file in your FastAPI backend, its actual extension or content might not match what the Whisper API expects—even if other browsers produce a proper wav file.

Recommendations

1. Inspect the Actual File Format:
   Check the real content and file extension of the blob that Safari produces. For example, log or inspect the file (e.g. using a file inspector tool or even a simple file header check in Python) to verify its format.

2. Ensure Correct File Extension:
   When you create the temporary file in FastAPI, make sure its filename (and extension) reflect the underlying audio format. You might need to extract or enforce the correct extension (e.g. .m4a if that’s what the encoder is producing).

3. Conversion on the Backend:
   If you can’t control what Safari produces, consider using a conversion tool (like ffmpeg) on the backend to convert the file to a supported format that matches the declared MIME type before sending it to the Whisper API.

4. Testing with Multiple MIME Types:
   You may also try setting the MIME type to match what Safari is actually using (for example, audio/mp4 or audio/m4a) when creating the blob from the recording.

Example FastAPI Endpoint

Below is an example of how you might inspect the file extension and enforce a proper temporary filename:

PYTHONCopied
from fastapi import FastAPI, File, UploadFile, HTTPException
import tempfile
import shutil

app = FastAPI()

ALLOWED_EXTENSIONS = {"m4a", "mp3", "webm", "mp4", "mpga", "wav", "mpeg"}

@app.post("/audio")
async def process_audio(file: UploadFile = File(...)):
    # Extract the extension from the uploaded file name
    original_filename = file.filename
    extension = original_filename.rsplit(".", 1)[-1].lower() if "." in original_filename else None
    if extension not in ALLOWED_EXTENSIONS:
        raise HTTPException(
            status_code=400,
            detail=f"Invalid file format. Supported formats: {sorted(ALLOWED_EXTENSIONS)}"
        )
    
    # Write the uploaded file to a temp file with a proper extension
    with tempfile.NamedTemporaryFile(delete=False, suffix=f".{extension}") as tmp:
        shutil.copyfileobj(file.file, tmp)
        temp_file_path = tmp.name

    # Now call the Whisper API with temp_file_path
    # For example: result = call_whisper_api(temp_file_path)

    return {"detail": "File received and processed successfully."}

Summary

Safari on iOS might be producing an audio file with a different container or file extension than expected—even if you set the MIME type manually. Investigate the actual file details and adjust the temporary file naming or perform conversion if necessary, so that the file matches one of the allowed formats by the Whisper API.