Add support for passing buffers to transcribe about faster-whisper

[guillaumekln]

Sure, that should be a simple change. Let me open a PR.

[guillaumekln]

Hi,

This is already possible. The transcribe method accepts file-like objects so you can wrap your buffer with io.BytesIO.

For example, the code below reads the audio file into a buffer of bytes which is then passed to transcribe:

import io

with open("audio.mp3", "rb") as audio_file:
    audio_bytes = audio_file.read()
    audio_file = io.BytesIO(audio_bytes)

segments, _ = model.transcribe(audio_file)
text = "".join(segment.text for segment in segments)

[royce-mathew]

Would it be possible to add support to pass the waveform as a numpy array directly to transform without creating a file? whisper currently allows you to pass the audio file or the audio waveform. For projects which listen to audio in realtime and try to transcribe the audio fast I think the option of being able to pass the waveform directly to transcribe would be really great