Recherche avancée

Sur d’autres sites (6678)

• How to improve my current method of generating an audio transcript from an online video

  23 janvier 2019, par 5areductase

  I am working on a project that needs to download and analyze online videos. Most of the videos are hosted on YouTube but some are hosted on Facebook.

  My task is to generate an audio transcript for every video.

  My colleague was using a series of programs sequentially given some {link} :

  youtube-dl -f '(mp4)[height = 360][width = 640]' {link} -o '{out_1}.%(ext)s'
  
  
  
  ffmpeg -i {out_1} -vn {out_2}.wav
  
  
  
  sox {out_2} {out_3} channels 1 rate 16000
  
  
  
  pocketsphinx_continuous -infile {out_3} -samprate 16000 -hmm {ACOUSTICMODEL} -dict {DICTIONARY} -lm {LANGMODEL} -fwdflat yes -bestpath yes 2> error.log | tee {out_4}.log && rm error.log

  Note that there’s an extra ffmpeg step to extract the audio, instead of simply directly downloading it with youtube-dl because video is needed as well.

  Everything works correctly as far as I can tell, but I’ve never really dealt with audio before so I’m not sure if this is the best way to go about it.

• ffmepg video from uneven sequence of png images

  26 mai 2015, par Maxim Kornienko

  I’m going to make a video from series of screenshots (.png files). For each screenshot there is assosiated timestamp information about when it was taken. The time intervals between screenshots may vary and it’s highly desired to preserve that difference in the video.

  Is there a way to use single ffmpeg command/API providing to it sequence of image + time (or frame) offset and getting one video file as output ?
  By now I have to generate short video files of custom length for each image, and then merge them manually :

  ffmpeg -y -loop 1 -i image1.png -c:v libx264 -t 1.52 video1.avi
  
  ffmpeg -y -loop 1 -i image2.png -c:v libx264 -t 2.28 video2.avi
  
  ...
  
  ffmpeg -y -loop 1 -i imageN.png -c:v libx264 -t 1.04 videoN.avi
  
  
  
  
  
  ffmpeg -i "concat:video1.avi|video2.avi|...videoN.avi" -c copy output.avi

  This is quite ok, while intervals are large, but the whole approach seems to me a bit fragile.

• Converting RGB images into lossless video with avconv / ffmpeg [migrated]

  19 mai 2013, par despens

  I am trying to convert a series of PNG24 images containing a screen animation to a lossless video, so that every pixel is reproduced exactly as it was in the original. But avconv (and ffmpeg) both produce the same, unsatisfying results, which seem to stem from a colospace conversion going wrong.

  This is the version of avconv I am using :

  avconv version 0.8.6-4:0.8.6-0ubuntu0.12.04.1, Copyright (c) 2000-2013 the Libav developers
  
    built on Apr  2 2013 17:02:36 with gcc 4.6.3

  Each of the images is 1280×800 pixels in size. They do not contain any photographic motives, but specially dithered patterns.

  I used the qtrle codec because apparently this is a lossless codec that works very much like animated GIFs or animated PNGs. However, during the conversion to video there seems to happen a color space conversion ("filter") that is messing with the pixels.

  This is the avconv output :

  $ avconv -f image2 -r 30 -i frames.png/wth-%08d.png -vcodec qtrle -pix_fmt rgb24 -t 15 qtrle-30fps-rgb.mov
  
  avconv version 0.8.6-4:0.8.6-0ubuntu0.12.04.1, Copyright (c) 2000-2013 the Libav developers
  
    built on Apr  2 2013 17:02:36 with gcc 4.6.3
  
  Input #0, image2, from 'frames.png/wth-%08d.png':
  
    Duration: 00:11:17.96, start: 0.000000, bitrate: N/A
  
      Stream #0.0: Video: png, bgra, 1280x800, 30 fps, 30 tbr, 30 tbn, 30 tbc
  
  File 'qtrle-30fps-rgb.mov' already exists. Overwrite ? [y/N] y
  
  [buffer @ 0x740a00] w:1280 h:800 pixfmt:bgra
  
  [avsink @ 0x742ac0] auto-inserting filter 'auto-inserted scaler 0' between the filter 'src' and the filter 'out'
  
  [scale @ 0x743680] w:1280 h:800 fmt:bgra -> w:1280 h:800 fmt:rgb24 flags:0x4
  
  Output #0, mov, to 'qtrle-30fps-rgb.mov':
  
    Metadata:
  
      encoder         : Lavf53.21.1
  
      Stream #0.0: Video: qtrle, rgb24, 1280x800, q=2-31, 200 kb/s, 30 tbn, 30 tbc
  
  Stream mapping:
  
    Stream #0:0 -> #0:0 (png -> qtrle)
  
  Press ctrl-c to stop encoding
  
  frame=  450 fps= 22 q=0.0 Lsize=  126056kB time=15.00 bitrate=68843.4kbits/s    
  
  video:126052kB audio:0kB global headers:0kB muxing overhead 0.003441%

  This is an original PNG image file :

  

  This is a screenshot of the video being replayed :

  

  Please note that you need to see these images in their original 1280×800 resolution to notice the differences.

  Here is a side-by-side comparison, with the left picture being the original and the right one the outcome after video encoding :

  

  Is there any way I can produce a truly lossless, pixel-perfect video file from a series of PNGs ?