The ecapa verification works well but quite slow (~7 mins for 100 pairs).
Consider running inference on gpu if the machine has one.
add this argument run_opts={"device":"cuda"} to the code below. Also make sure that the audio tensors are copied to the device as well e.g. audio_tensor.to("cuda").
The ecapa verification works well but quite slow (~7 mins for 100 pairs).
Consider running inference on gpu if the machine has one.
add this argument
run_opts={"device":"cuda"}
to the code below. Also make sure that the audio tensors are copied to the device as well e.g.audio_tensor.to("cuda")
.https://github.com/sensein/b2aiprep/blob/85dcd1146b305dc0f965d0abf3a7ab810088ca58/src/b2aiprep/process.py#L64
Source: https://huggingface.co/speechbrain/spkrec-ecapa-voxceleb#inference-on-gpu