Recherche avancée

Sur d’autres sites (10641)

• Extracting subtitles of a mkv file using ffmpeg while downloading

  19 août 2020, par Tim Untersberger

  I am trying to stream a mkv file to a webpage, which is being downloaded using WebTorrent. The video file has ASS/SSA embedded subtitles. I am using electron to display the video, which uses chromium without proprietary codec support, so I have to use the following html :

  


  <video controls="controls">&#xA;  <source src="video.mkv" type="video/webm">&#xA;<video></video>&#xA;</source></video>

  


  Since chrome supports webm and they use the same container, I can watch the video with audio. Webm doesn't support embedded subtitles, so the subtitles of the video get ignored.

  


  I am extracting the subtitles using the following command :

  


  ffmpeg -i video.mkv -map 0:2 sub.vtt


  


  This works fine, but requires the file to be downloaded completely. Is it possible to extract the subtitles while the file is being downloaded, so that I can begin streaming the video together with the subtitles as soon as possible ?

  


• ffmpeg - how to handle video and audio URL's streams separately ?

  13 mai 2022, par Rubem Pacelli

  I am trying to create a zsh function that uses youtube-dl and ffmpeg to download a portion of a YouTube video. I did achieve this goal with the following function :

  


  # $1 - youtube URL
# $2 - start position in hh:mm:ss.msms format (ms=miliseconds)
# $3 - final position in hh:mm:ss.msms format (ms=miliseconds)
# $4 - output file name (optional)
function youtubedl_snippet()(
  local url_stream=$(youtube-dl -f best --get-url $1)
  local output_name=$(youtube-dl --get-title $1)

  ffmpeg -ss $2 -to $3 -i $url_stream -c:v copy -c:a copy ${4:-"$output_name.mp4"}
)


  


  The command youtube-dl -f best --get-url $1 return a single URL with the best possible quality. In order to understand better how ffmpeg works, I tried to create another function with the same goal but with a different approach :

  


  # $1 - youtube URL
# $2 - start position in hh:mm:ss.msms format (ms=miliseconds)
# $3 - final position in hh:mm:ss.msms format (ms=miliseconds)
# $4 - output file name (optional)
# $5 - output video codec type (optional, for instance: libx264)
# $6 - output audio codec type (optional, for instance: aac)
function youtubedl_snippet2()(
  local url_streams=$(youtube-dl --get-url $1)
  local output_name=$(youtube-dl --get-title $1)

  local url_array=(${(f)url_streams}) # expand urls by lines url_array[1] -> video stream url_array[2] -> audio stream

  ffmpeg -ss $2 -to $3 -i ${url_array[1]} -ss $2 -to $3 -i ${url_array[2]} -map 0:v -map 1:a -c:v ${5:-copy} -c:a ${6:-copy} ${4:-"$output_name.mp4"}
)


  


  What I suppose that is going on here :

  


  

  1. url_streams is a line-separated URL. url_array[1] is the video URL and url_array[2] is the audio URL.

  


  2. I am setting both audio and video to the same intervals.

  


  3. I mapped the first stream to video, and the second to audio

  


  4. If $5 and $6 does not give different codecs, ffmpeg just copy from the original source.

  


  


  Well, it seems that everything is ok. But when I try

  


  start=$SECONDS; youtubedl_snippet2 'https://www.youtube.com/watch?v=g-_hVXzkn0o' 00:00:05.00 00:00:15.00; echo "it takes $(( SECONDS - start )) seconds"


  


  It will take 368 seconds. Moreover, I cannot open it in my android (only audio works)

  


  

  


  On the other hand,

  


  start=$SECONDS; youtubedl_snippet 'https://www.youtube.com/watch?v=g-_hVXzkn0o' 00:00:05.00 00:00:15.00; echo "it takes $(( SECONDS - start )) seconds"


  


  takes only 40 seconds, and the video can be played on Android.

  


  Here is the youtubedl_snippet log file. And here is the youtubedl_snippet2 log file.

  


• x86/tx_float : remove vgatherdpd usage

  20 mai 2022, par Lynne

  x86/tx_float : remove vgatherdpd usage
  Its performance loss ranges from either being just as fast as individual loads
  
  (Skylake), a few percent slower (Alderlake), 8% slower (Zen 3), to completely
  
  disasterous (older/other CPUs).
  Sadly, gathers never panned out fast on x86, even with the benefit of time and
  
  implementation experience.
  This also saves a register, as there's no need to fill out an additional
  
  register mask.
  Zen 3 (16384-point transform) :
  
  Before : 1561050 decicycles in           av_tx (fft),  131072 runs,      0 skips
  
  After :  1449621 decicycles in           av_tx (fft),  131072 runs,      0 skips
  Alderlake :
  
  2% slower on big transforms (65536), to 1% (131072), to a few percent for smaller
  
  sizes.
  

  • [DH] libavutil/x86/tx_float.asm
  • [DH] libavutil/x86/tx_float_init.c