First, thanx for the extrem usefull software/work. Very nice, but in diarize_parallel.py line 166 please replace this code:
emissions, stride = generate_emissions(
alignment_model,
audio_waveform.to(alignment_model.dtype).to(alignment_model.device),
batch_size=args.batch_size,
)
with this from diarize.py
emissions, stride = generate_emissions(
alignment_model,
torch.from_numpy(audio_waveform)
.to(alignment_model.dtype)
.to(alignment_model.device),
batch_size=args.batch_size,
)
And diarize_parallel.py will working with cuda, extremly fast.