Whisper medium performing differently through cli and library

[swap-nil7]

We tried to use whisper-medium for our use case since when tested on cli it was giving very nice results. We integrated it using the library as shown

import whisper

model = whisper.load_model("base")
result = model.transcribe("audio.mp3")
print(result["text"])

But now the results are not the same, please let us know if the results are different through cli and library approaches

[glangford]

See #177