import cv2
import torch
import datasets
from torchcodec.decoders import AudioDecoder
from torchcodec.decoders import VideoDecoder
def load_audio(source:str|bytes, start_time:int=0, end_time:int|None=None):
audio_decoder = AudioDecoder(source)
if end_time is None:
end_time = audio_decoder.metadata.duration_seconds_from_header
waveform = audio_decoder.get_samples_played_in_range(start_time, end_time).data
return waveform.transpose(1, 0) # T x 1… See the full description on the dataset page:
https://huggingface.co/datasets/MahmoodAnaam/LRS2-Pretrain.
