Algorithm info: Submitted by Bas Curtiz
I first tried the 3-day trial, but LANDR Stems wasn't available.
Then mailed them to ask what the cheapest way was.
They replied it's only available in non-trial.
I subbed for Essential non-trial, but nope, still I couldn't use LANDR Stems.
LANDR Stems is only included in Studio Standard + Studio Pro, it turned out.
They were so kind to upgrade my account to Standard and apologized for the confusion.
Output:
32-bit WAV
No batch-mode
4-stems only, but merged bass/drums/other together to form the instrumental part with:
import soundfile as sf
import numpy as np
import os
from tqdm import tqdm
def process(path: str):
# Load audio files
files = []
for f in os.listdir(path):
file = os.path.join(path, f)
data, sr = sf.read(file)
files.append((data, sr, f))
# Sum audio data to create mixture
mixture = sum([i[0] for i in files])
# Write individual files and the mixture
for (data, sr, file) in files:
sf.write(os.path.join(path, os.path.splitext(file)[0] + '_FIXED.wav'), data, sr, subtype='PCM_32')
sf.write(os.path.join(path, '_MIXTURE.wav'), mixture, files[0][1], subtype='PCM_32')
if __name__ == '__main__':
for root, folders, _ in tqdm(os.walk('.')):
for folder in folders:
folder = os.path.join(root, folder)
if len([f for f in os.listdir(folder) if f.endswith('.wav')]) == 0:
continue
process(folder)
Metrics:
Metric sdr for instrum: 12.6435
Metric si_sdr for instrum: 12.5064
Metric l1_freq for instrum: 28.7288
Metric log_wmse for instrum: 10.9344
Metric aura_stft for instrum: 7.5736
Metric aura_mrstft for instrum: 9.9870
Metric bleedless for instrum: 36.9348
Metric fullness for instrum: 19.1035
Metric sdr for vocals: 9.1909
Metric si_sdr for vocals: 8.4915
Metric l1_freq for vocals: 34.4358
Metric log_wmse for vocals: 12.4829
Metric aura_stft for vocals: 6.8718
Metric aura_mrstft for vocals: 7.9362
Metric bleedless for vocals: 32.5553
Metric fullness for vocals: 13.5451
Date added: 2024-12-09 |
