Speaker diarization correctly identifies speakers in notebook but not when run on fastapi endpoint

[ColtonBehannon]

I have two versions of a program with the same intention for both: speaker diarization. One of the files is a python file that hosts a fastapi endpoint that can be curled to to diarize an audio file. The other is a .ipynb that contains the exact same code except the fastapi endpoint definition has been removed (should not affect anything) and the audio file is being passed differently. Here are the differences below:

.py

@app.post("/")
async def api_create_order(request: Request):
    data = await request.body()

    temp = NamedTemporaryFile()

    dest_file = open(temp.name, 'wb+')
    dest_file.write(data)
    dest_file.close()

diarization = pipeline(temp.name, num_speakers=2)

With curl -F "file=@<filepath>" localhost:5000/diarize being used to pass the file

.ipynb

data = '<filepath>'
temp = NamedTemporaryFile()
with open(data, "rb") as f:
    data_bytes = f.read()

dest_file = open(temp.name, 'wb+')
dest_file.write(data_bytes)
dest_file.close()

diarization = pipeline(temp.name, num_speakers=2)

This is the only difference between the two files and they are run on the same kernels. The results for the .ipynb are very good and identify the speakers swapping back and forth throughout the audio. The .py however only identifies one speaker for 99% of the file and then finally identifies the other speaker at the very end of the audio. What am I missing? Any help appreciated.

[ColtonBehannon]

Seemed to be an issue of passing the data as a request body versus a file upload. The hash was coming through differently when passed a request body. The transcription from whisper worked fine on the same file so it was strange that the speaker identification was struggling. Regardless, passing it as a UploadFile solved it.