How to use whisperX to do forced alignment

[jhkonan]

I understand whisper is able to give the predicted transcript and alignment. However, I have both the audio and the original transcript, so I would like to do forced alignment with whisper (turbo) using whisperX.

Is this capability possible within the current framework? If a working example could be made, it would be very helpful.

Here is an example speech text pair to start...

!wget https://catalog.ldc.upenn.edu/desc/addenda/LDC93S1.wav -O LDC93S1.wav
!wget https://catalog.ldc.upenn.edu/desc/addenda/LDC93S1.txt -O LDC93S1.txt

import librosa
import numpy

speech, sr = librosa.load("LDC93S1.wav", sr=16000)
text = " ".join(numpy.loadtxt("LDC93S1.txt", dtype=str)[2:])

[zou8944]

you can refer this: https://huggingface.co/ericmattmann/whisperX-endpoint/blob/961372e2ea1e0fa0dcf34d9d797e9d3dffabdd6d/handler.py#L281