Recherche avancée

Sur d’autres sites (5191)

• Unable to open audio file on Heroku using Librosa

  15 mars 2020, par Rohan Bojja

  I have a feature extraction REST API written in Python using the Librosa library (Extracting audio features), it receives an audio file through HTTP POST and responds with a list of features(such as MFCC,etc).

  Since librosa depends on SoundFile (libsndfile1 / libsndfile-dev), it doesn’t support all the formats, I’m converting the audio file using ffmpeg-python wrapper (https://kkroening.github.io/ffmpeg-python/) .

  It works just fine on my Windows 10 machine with Conda, but when I deploy it on Heroku, the librosa.load() functions returns an unknown format error, no matter what format I convert it to. I have tried FLAC, AIFF and WAV.

  My first guess is that the converted format isn’t supported by libsndfile1, but it works on my local server (plus, their documentation says AIFF and WAV are supported), so I’m a little lost.

  I have attached all the relevant snippets of code below, I can provide anything extra if necessary. Any help is highly appreciated. Thanks.

  UPDATE1 :

  I am using pipes instead of writing and reading from disk, worth a mention as the question could be misleading otherwise.

  The log :

  File "/app/app.py", line 31, in upload
  
  x , sr = librosa.load(audioFile,mono=True,duration=5)
  
  File "/app/.heroku/python/lib/python3.6/site-packages/librosa/core/audio.py", line 164, in load
  
  six.reraise(*sys.exc_info())
  
  File "/app/.heroku/python/lib/python3.6/site-packages/six.py", line 703, in reraise
  
  raise value
  
  File "/app/.heroku/python/lib/python3.6/site-packages/librosa/core/audio.py", line 129, in load
  
  with sf.SoundFile(path) as sf_desc:
  
  File "/app/.heroku/python/lib/python3.6/site-packages/soundfile.py", line 629, in __init__
  
  self._file = self._open(file, mode_int, closefd)
  
  File "/app/.heroku/python/lib/python3.6/site-packages/soundfile.py", line 1184, in _open
  
  "Error opening {0!r}: ".format(self.name))
  
  File "/app/.heroku/python/lib/python3.6/site-packages/soundfile.py", line 1357, in _error_check
  
  raise RuntimeError(prefix + _ffi.string(err_str).decode('utf-8', 'replace'))
  
  RuntimeError: Error opening <_io.BytesIO object at 0x7f46ad28beb8>: File contains data in an unknown format.
  
  10.69.244.94 - - [15/Mar/2020:12:37:28 +0000] "POST /receiveWav HTTP/1.1" 500 290 "-" "curl/7.55.1"

  Flask/Librosa code deployed on Heroku (app.py) :

  from flask import Flask, jsonify, request
  
  import scipy.optimize
  
  import os,pickle
  
  import numpy as np
  
  from sklearn.preprocessing import StandardScaler
  
  import librosa
  
  import logging
  
  import soundfile as sf
  
  from pydub import AudioSegment
  
  import subprocess as sp
  
  import ffmpeg
  
  from io import BytesIO
  
  
  
  logging.basicConfig(level=logging.DEBUG)
  
  
  
  app = Flask(__name__) 
  
  
  
  @app.route('/receiveWav',methods = ['POST'])
  
  def upload():
  
      if(request.method == 'POST'):
  
          f = request.files['file']
  
          app.logger.info(f'AUDIO FORMAT\n\n\n\n\n\n\n\n\n\n: {f}')
  
          proc = (
  
              ffmpeg.input('pipe:')
  
              .output('pipe:', format='aiff')
  
              .run_async(pipe_stdin=True,pipe_stdout=True, pipe_stderr=True)
  
          )
  
          audioFile,err = proc.communicate(input=f.read())
  
          audioFile =  BytesIO(audioFile)
  
          scaler = pickle.load(open("scaler.ok","rb"))
  
          x , sr = librosa.load(audioFile,mono=True,duration=5)
  
          y=x
  
          #Extract the features
  
          chroma_stft = librosa.feature.chroma_stft(y=y, sr=sr)
  
          spec_cent = librosa.feature.spectral_centroid(y=y, sr=sr)
  
          spec_bw = librosa.feature.spectral_bandwidth(y=y, sr=sr)
  
          rolloff = librosa.feature.spectral_rolloff(y=y, sr=sr)
  
          zcr = librosa.feature.zero_crossing_rate(y)
  
          rmse = librosa.feature.rms(y=y)
  
          mfcc = librosa.feature.mfcc(y=y, sr=sr)
  
          features = f'{np.mean(chroma_stft)} {np.mean(rmse)} {np.mean(spec_cent)} {np.mean(spec_bw)} {np.mean(rolloff)} {np.mean(zcr)}'    
  
          for e in mfcc:
  
              features += f' {np.mean(e)}'
  
          input_data2 = np.array([float(i) for i in features.split(" ")]).reshape(1,-1)
  
          input_data2 = scaler.transform(input_data2)
  
          return jsonify(input_data2.tolist())
  
  
  
  # driver function 
  
  if __name__ == '__main__':   
  
      app.run(debug = True) 

  Aptfile :

  libsndfile1
  
  libsndfile-dev
  
  libav-tools
  
  libavcodec-extra-53 
  
  libavcodec-extra-53
  
  ffmpeg 

  requirements.txt :

  aniso8601==8.0.0
  
  audioread==2.1.8
  
  certifi==2019.11.28
  
  cffi==1.14.0
  
  Click==7.0
  
  decorator==4.4.2
  
  ffmpeg-python==0.2.0
  
  Flask==1.1.1
  
  Flask-RESTful==0.3.8
  
  future==0.18.2
  
  gunicorn==20.0.4
  
  itsdangerous==1.1.0
  
  Jinja2==2.11.1
  
  joblib==0.14.1
  
  librosa==0.7.2
  
  llvmlite==0.31.0
  
  MarkupSafe==1.1.1
  
  marshmallow==3.2.2
  
  numba==0.48.0
  
  numpy==1.18.1
  
  pycparser==2.20
  
  pydub==0.23.1
  
  pytz==2019.3
  
  resampy==0.2.2
  
  scikit-learn==0.22.2.post1
  
  scipy==1.4.1
  
  six==1.14.0
  
  SoundFile==0.10.3.post1
  
  Werkzeug==1.0.0
  
  wincertstore==0.2

• Clarification on using FFmpeg in an open source app with —enable-gpl and —enable-nonfree [closed]

  24 avril 2020, par Alex Hoffman

  Can someone please clarify the legality of bundling and using the full version of FFmpeg (with included libx264 and libx265 and other gpl codecs) in an open source app.

  



  I don't want to modify anything, all I want to do is to bundle compiled FFmpeg binaries downloaded from their website with my open source app which would use them for media files editing.

  



  I want to bundle the full version of FFmpeg since without --enable-gpl and --enable-nonfree, not being able to process h264, FFmpeg becomes effectivelly useless. So I want to know if I have to bundle the binaries compiled without --enable-gpl and --enable-nonfree or can I bundle the full version with shared linking (libs separated from the executable) ?

  


• ffmpeg avformat_open_input() failed to open a dshow device url containing Chinese Characters

  16 octobre 2024, par Sang

  The USB Webcam is OK, but the device I want to use is a "virtual camera" named "无他伴侣(竖屏)", whose video is from the smart phone like Android or iOS. Connect the phone to PC, run an app on the phone, then run a PC client application, which can preview the video. The phone app is called "无他相机", and the PC app is called "无他伴侣", its website is https://www.wuta-cam.com/.

  


  I run FFmpeg at the Windows Commandline with the command ffmpeg -list_devices true -f dshow -i dummy, it's OK to list the devices.(in order to display Chinese correctly, run chcp 65001 in advance.)

  


  Run the command ffplay -f dshow -i video="无他伴侣(竖屏)", it's OK to play the video.(of course you need to comfirm its PC client previewing fine in advance.)

  


  Now I want to get the decoded frames from that virtual camera in my program, I call avformat_open_input() with video=无他伴侣(竖屏), it failed, the return value is -5, I/O error.

  


  Anyone knows the reason ?

  


  Below is my code snippet.

  


  avdevice_register_all();
avcodec_register_all();
//const char * url= "video=Logitech Webcam C930e";// This is fine.
char url[] = "video=无他伴侣(竖屏)";// This is bad.

AVFormatContext *pFmtCtx = avformat_alloc_context();
AVInputFormat *iformat = av_find_input_format("dshow");
int nRet = 0;
nRet = avformat_open_input(&pFmtCtx, url, iformat, NULL);
if (nRet)
{
    const size_t buffer_size = 256;
    char err_description[buffer_size];
    av_strerror(nRet, err_description, buffer_size);
    printf("%s.\n", err_description);// --> I/O error.
    printf("FAILED to open input.(Line:%d,%d)\n",  __LINE__, nRet);
    return -1;
}