Recherche avancée

Sur d’autres sites (4928)

• Failed to hard-ware accelerate video decoding via Tesla P40

  20 mai 2018, par Potemkin

  While I am writing a surveillance video recognition demo, I find it is much slower to simply play a video in the server(Xeon E5-2680 2.4GHz, Tesla P40) than in my laptop(i7-8550 1.8GHz, Intel UHD Graphics 620).

  I use DXVA-Checker to see the video decoder device, and notice that my laptop use the GPU for decoding, but the server use no GPU for the job. Then I check ’nvidia-smi’, and it shows all of GPUs are running in TCC driver model, not WDDM driver model. And I cannot turn it to WDDM because nvidia-smi says it is not supported.

  The video play demo is written via OPENCV, in which the video decoding part is from ffmpeg. The server runs in Windows-server-2012, and my laptop is on Windows 10.

  The question is how can I get the server decoding videos with GPUs, and is this the reason for the slowness or there is something else ?

• demucs.api or demucs.pretrained aren't working

  9 décembre 2023, par user22801597

  I created my environment and installed several packages for CUDA 11.8.0 (including demucs, pytorch, torchaudio, ffmpeg...). I want to separate a track with demucs with a python code in spyder (i don't want to do it on the anaconda prompt) however I've tried several codes and they seem not be working. I get errors like :

  


  

  • AttributeError : module 'demucs' has no attribute 'Separator' (1)

  


  • ImportError : cannot import name 'load_pretrained' from 'demucs.pretrained' (2) (C :\Users\User\anaconda3\envs\Pytorch\Lib\site-packages\demucs\pretrained.py)

  


  • ModuleNotFoundError : No module named 'demucs.api' (3)

  


  • ModuleNotFoundError : No module named 'demucs.model'(4)

  


  


  So through the errors I assume my demucs package isn't working correctly but I don't know why. I've done the following (I'Ve found this either in google or in chatgpt) :

  


  

  • Checked Library Versions

  


  • Upgraded demucs and other packages
-deinstalled and reinstalled it
Can someone please help me to solve this ? What am I doing worng or why is demucs not working correctly ?

  


  


  This are the different codes I tested according to their errors above :

  


  (1)

  


     import demucs
   separator = demucs.Separator()
   model = demucs.load_pretrained('demucs')
   audio_sources = separator.separate('Zweite_Aufnahme_mit_Piano.wav')
   for source_name, source_audio in audio_sources.items():
       source_audio.to_file(source_name + '.wav')



  


  (2)

  


     import torchaudio
   from demucs.pretrained import load_pretrained
   import soundfile as sf
   import os

   input_audio_file = 'Zweite_Aufnahme_mit_Piano.wav'
   output_directory = 'Separated_audios'

   model = load_pretrained('demucs_quantized')

   audio, _ = torchaudio.load(input_audio_file)

   sources = model(audio)

   os.makedirs(output_directory, exist_ok=True)

   for i, source in enumerate(sources):
       source_path = os.path.join(output_directory, f'source_{i}.wav')
       sf.write(source_path, source[0].numpy(), 44100)

   print("Source separation complete. Separated sources saved in the output directory.")



  


  (3)

  


     from pydub import AudioSegment
   import IPython
   import numpy as np
   import demucs.api
   import pathlib
   import torch

   audiosegment = AudioSegment.from_mp3("Zweite_Aufnahme_mit_Piano.wav")
   arr = audiosegment.get_array_of_samples()
   sr = audiosegment.frame_rate
   ch = audiosegment.channels
   aac = np.array(arr, dtype=np.float32).reshape((-1, ch))
   aac = aac[30*sr:40*sr,:]
   IPython.display.Audio(aac[:, 0], rate=sr)

   separator = demucs.api.Separator()
   model = separator.load_model()
   input = torch.from_numpy(aac.reshape((ch, -1)))
   separator.add_track("born", input)
   separated = separator.separate_loaded_audio()
   for (file, stems) in separated:
       for stem, waveform in stems.items():
         print(stem)
         print(waveform.shape)
         aac_again = waveform.numpy().reshape((-1, ch))
         IPython.display.display(IPython.display.Audio(aac_again[:,0], rate=sr))



  


  (4)

  


     import torchaudio
   from demucs.model import load_pretrained

   audio, sample_rate = torchaudio.load('Zweite_Aufnahme_mit_Piano.wav')
   model = load_pretrained('demucs')
   separated_sources = model(audio)
   torchaudio.save('vocals.wav', separated_sources['vocals'], sample_rate)
   torchaudio.save('accompaniment.wav', separated_sources['accompaniment'], sample_rate)
   torchaudio.save('drums.wav', separated_sources['drums'], sample_rate)
   torchaudio.save('bass.wav', separated_sources['bass'], sample_rate)


  


• Anomalie #3318 (Fermé) : Problème sur le filtre extraire_multi

  31 octobre 2014, par cedric -

  Corrigé, le patch etait inexact car il faut bien typographier en fr dans le cas ou on prend fr au lieu de la langue courante