Spaces:
Sleeping
Sleeping
File size: 1,313 Bytes
541c6b7 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 |
import argbind
import audiotools as at
import numpy as np
import torch
import tqdm
import dac
@argbind.bind(without_prefix=True, positional=True)
def main(
folder: str,
model_path: str,
n_samples: int = 1024,
device: str = "cuda",
):
files = at.util.find_audio(folder)[:n_samples]
signals = [
at.AudioSignal.salient_excerpt(f, loudness_cutoff=-20, duration=1.0)
for f in files
]
with torch.no_grad():
model = dac.model.DAC.load(model_path).to(device)
model.eval()
codes = []
for x in tqdm.tqdm(signals):
x = x.to(model.device)
o = model.encode(x.audio_data, x.sample_rate)
codes.append(o["codes"].cpu())
codes = torch.cat(codes, dim=-1)
entropy = []
for i in range(codes.shape[1]):
codes_ = codes[0, i, :]
counts = torch.bincount(codes_)
counts = (counts / counts.sum()).clamp(1e-10)
entropy.append(-(counts * counts.log()).sum().item() * np.log2(np.e))
pct = sum(entropy) / (10 * len(entropy))
print(f"Entropy for each codebook: {entropy}")
print(f"Effective percentage: {pct * 100}%")
if __name__ == "__main__":
args = argbind.parse_args()
with argbind.scope(args):
main()
|