Translation(s): English - French
FFmpeg/Libav is a suite of open source software in command line that permits managing of audio or video streams. With FFmpeg/Libav, you can make recordings, corrections with filters, or transcode media from one format to an other.
Libav is a fork of FFmpeg that can be found in the Debian repositories. If you want to use ffmpeg, you will need to change the word "avconv" to "ffmpeg" in the different commands lines. The rest stay the same.
Contents
Installation
To install libav:
apt-get install libav-tools
To install ffmpeg:
apt-get install ffmpeg
Be careful, FFmpeg is not available on Debian 8 Jessie, but in Backports. To find out which deposit the package is available, look here: debpkg:libav or here debpkg:ffmpeg . It also possible to compile it by following the instructions available here: https://trac.ffmpeg.org/wiki/CompilationGuide/Ubuntu .
Usecases
FFmpeg or Libav can be use from GUI like:
Or directly from the command line interface, as explained bellow.
Introduction
An introduction to the terminology of video is useful to understand the FFmpeg syntax.
Formats
The format is the container that allows the transport of video, audio and subtitles or as a file (mkv, mov ...) or via stream (MPEG TS). Inside a container can be inserted (mux) or extract (demux):
- one or more video streams (a movie or tv channels...)
- one or more sound streams (VO, VF, ...)
- one or more streams of subtitles. (English, hearing impaired, ...)
- plus metadatas (title, artist's name, ....)
We talk about multiplexing different tracks or streams in a format.
FFmpeg/Libav provide a list of supported format:
avconv -formats
... DE avi AVI (Audio Video Interleaved) DE ogg Ogg D matroska,webm Matroska / WebM E mov QuickTime / MOV D mov,mp4,m4a,3gp,3g2,mj2 QuickTime / MOV E webm WebM ...
- D. = Demuxing supported
- .E = Muxing supported
Codecs
The codec is an algorithm used to encode the video or sound to suit the transport protocol (IP, DVB, file ...) including reducing the flow rate (Kbits / s). By using codecs, compression can result in the loss of perceivable quality in picture or sound.
In the same way as for formats, FFmpeg / Libav lists the codecs that it can handle:
avconv -codecs
Filters
FFmpeg / Libav also has a significant base of filters that can modify the content of each stream, such as changing resolution, change the volume of a track, superimpose a logo etc ....
avconv -filters
Formats + Codecs + Filters
To summarize, FFmpeg / Libav can multiplex or de-multiplex in different formats:
- Compressed video streams (or not)
- Compressed audio stream (or not)
- Subtitles in various formats.
And modify, using filters, the content of each stream independently.
But if FFmpeg / Libav can handle tons of formats and tons of different codecs, all combinations are not possible, as shown in this table: https://www.videolan.org/streaming-features.html.
Indeed, each codec or format has its own standard with more or less restrictive licenses, all combinations are not possible. Fortunately the free world provides several formats as well as several free codecs:
- In formats: mkv, webm, ogv, ogg ...
- In the video codecs: VP9, VP8, theora (vp3) dirac ...
- In the audio codecs: flac, opus, vorbis, and many other ...
Knowing the contents of a file
Before starting any encoding it is good to know its contents, ffprobe or avprobe can read the header of the "format":
avprobe tears_of_steel.mkv
Which return:
Input #0, matroska,webm, from 'tears_of_steel.mkv': Metadata: title : ARTIST : COMPOSER : SYNOPSIS : DATE_RELEASED : GENRE : ENCODER : Lavf54.29.104 Duration: 00:12:14.12, start: 0.000000, bitrate: 4615 kb/s Stream #0:0(eng): Video: h264 (Main), yuv420p, 1280x534 [SAR 1:1 DAR 640:267], 24 fps, 24 tbr, 1k tbn, 180k tbc (default) Stream #0:1(eng): Audio: aac, 44100 Hz, stereo, s16 (default) Stream #0:2(eng): Audio: ac3, 48000 Hz, 5.1(side), s16, 448 kb/s (default) Stream #0:3(eng): Subtitle: ssa (default) Stream #0:4(fr): Subtitle: ssa (default)
So for this file, we can see that:
- The input 0 describes the file format (Matroska here). (Some formats allow multiple input, hence the numbering)
- The file includes 5 streams numbered from 0 to 4:
- an encoded video stream in h264
- an audio stream in English stereo
- a second audio stream in English 5.1
- and two subtitle one in English and one in French.
Change the Format
You can easily change format without affecting the stream:
avconv -i tears_of_steel.mkv -c copy tears_of_steel.mov
- -i specifies the input file
- -c copy: duplicate all streams without changing them
- -c: v Copy: duplicate the video tracks without changing them
- -c: a copy: duplicate identical audio tracks without changing them
- -c: s copy: duplicate identical tracks subtitles without changing them
If we do not tells ffmpeg/libav to make an exact copy of streams, it will launch an encoding default profile.
If we want to differentiate streams between them:
- -c: v: 0 first video track
- -c: a: 0 first audio track
- -c: a 1 second audio track
- -c s: 0 first track subtitle
- -c s: 2 second subtitle track
For files with more than 3 streams, it seems there is a bug with the "copy" function. Some streams are lost when copying, you have to use the mapping function to be able to copy all.
Organize the order of the streams (mapping)
Streams: _________ _________ _________ #0:0 | | | | | | +------->------->| decoder |--->| filters |--->| encoder |--->----+ | |_________| |_________| |_________| | __________ | _________ _________ _________ | ___________ | | | #0:1 | | | | | | | | | | Input #O |----+------->. ,-->| decoder |--->| filters |--->| encoder |--->----+----->| Output #O | |__________| | \ / |_________| |_________| |_________| | |___________| | X _______________________________________ | | #0:2 / \ | | | +------->ยด `-->| copy |--->----+ |_______________________________________| \________/ \______/ \________/ demuxer mapping muxer
ffmpeg/libav can change the order of the stream to suit your uses with the parameter " -map ":
avconv -i tears_of_steel.mkv -map 0:0 -map 0:2 -map 0:1 -map 0:4 -map 0:3 -c copy tears_of_steel-v2.mkv
In this command I inverted the two audio streams between them and also the subtitles.
- -map 0: 1: Track 1 of the input 0 (tears_of_steel.mkv)
- -map 0: 2: Track 2 of the input 0 (tears_of_steel.mkv)
When running, ffmpeg indicate which intersection (mapping) it applies:
Stream mapping: Stream #0:0 -> #0:0 (copy) Stream #0:2 -> #0:1 (copy) Stream #0:1 -> #0:2 (copy) Stream #0:4 -> #0:3 (copy) Stream #0:3 -> #0:4 (copy)
Add a stream
The parameter -map allow also to add a stream, in the following example a subtitle in Spanish (TOS-es.srt) in position 4:
avconv -i tears_of_steel.mkv -i TOS-es.srt -map 0:0 -map 0:1 -map 0:2 -map 1:0 -map 0:3 -map 0:4 -c:v copy -c:a copy -metadata:s:s:0 language=esp tears_of_steel-v2.mkv
- -map 0:0 refer to the first stream of the first "input": the video of the file "tears_of_steel.mkv"
- -map 1:0 refer to the first stream of the second "input": the subtitle of the file "TOS-es.srt"
This allows to insert it where you want:
- Stream #0:0(eng): Video: h264 (Main), yuv420p, 1280x534 [SAR 1:1 DAR 640:267], 24 fps, 24 tbr, 1k tbn, 180k tbc (default)
- Stream #0:1(eng): Audio: aac, 44100 Hz, stereo, s16 (default)
- Stream #0:2(eng): Audio: ac3, 48000 Hz, 5.1(side), s16, 448 kb/s (default)
- Stream #0:3(esp): Subtitle: ssa (default)
- Stream #0:4(eng): Subtitle: ssa (default)
- Stream #0:5(fr): Subtitle: ssa (default)
The option metadata can indicate the file language:
- -metadata:s:s:0 language=esp
Extract un stream
Using the possibilities of mapping it is very easy to extract one stream from a file:
avconv -i tears_of_steel.mkv -map 0:2 -acodec copy tears_of_steel_soundtrack.mkv
In this example, we extract just one audio track.
Extract a audio track in MP3
avconv -i tears_of_steel.mkv -map 0:2 -acodec libmp3lame -ar 44100 -ac 2 -ab 192k tears_of_steel_soundtrack.mp3
Extract a subtitle
avconv -i tears_of_steel.mkv -map 0:4 -c:s srt tears_of_steel_FR.srt
Extract a segment
FFmpeg/Libav can extract, pieces of media, stating a point of entry with '-ss '(-ss 00:06:46) and specifying a time with -t (-t 00:01:00). This function is very useful for making encoding tests to validate his command. Note that -ss and -t should be placed before the first input (-i).
avconv -ss 00:06:46 -t 00:01:00 -i tears_of_steel.mkv -c copy tears_of_steel_extract.mkv
Sometimes the term -t is not taken into account. You can use the option -to by placing it next to the output file
avconv -ss 00:06:46 -i tears_of_steel.mkv -c copy -to 00:01:00 tears_of_steel_extract.mkv
Increase the number of threads
Depending on the codec, it's possible to run multiple threads processors on a transcode with the option threads:
avconv -threads 4 -i tears_of_steel.mkv -c copy tears_of_steel_extract.mkv
Here I force it to work on 4 threads, that can save me valuable time. -threads 0 is the automatic mode and use the maximum of available threads, but it does not work with all the codecs.
Encoding videos
option |
details |
-b |
set bitrate (in bits/s) 200kbps by default |
-sameq |
use the bitrate of the source |
-s |
set the size of the image (-s 320x240) |
-aspect |
sample aspect ratio (4:3, 16:9 or 1.3333, 1.7777) |
-vcodec ou -c:v |
to choose a codec |
-pass |
Select the pass number, one pass (-pass 1) or two (-pass 2) |
-qmin |
minimum video quantizer scale (VBR) |
-qmax |
maximum video quantizer scale (VBR) |
-f |
Force input or output file format. |
-r |
set the frame rate |
Encode video in VP8
VP8 is a open source codec promoted by Google, a good replacement to H264/MP4
avconv -i tears_of_steel_720p.mkv -c:v:0 libvpx -crf 10 -vb 4M -c:a copy tears_of_steel_vp8.mkv
Here, we use the libvpx with two options:
- -crf defines a quality level between 0 and 63 (small number = better quality but more time to process)
- -vb 4M gives a target rate to 4 Megabit/s
A list of possible options for encoding in VP8 is available here:
Encode video in VP9
VP9 is a open source codec promoted by Google, to compete with h265/HEVC.
avconv -i tears_of_steel_720p.mkv -c:v:0 libvpx -crf 10 -vb 4M -c:a copy tears_of_steel_vp8.mkv
Here, we use the libvpx-vp9 with three options:
- -crf defines a quality level between 0 and 63 (small number = better quality but more time to process)
- -vb 4M gives a target rate to 4 Megabit/s
- -threads is necessary to force the number of core to use, libvpx-vp9 doesn't detect them yet.
A list of possible options for encoding in VP9 is available here:
vlc has an issue to decode VP9, see here: https://trac.videolan.org/vlc/ticket/11763 But it's possible to read it with ffplay:
ffplay tears_of_steel_vp9.mkv
Encode video in H264/Mpeg4
The H264/MPEG4 is a proprietary codec commonly used for its good quality/speed ratio:
avconv -i tears_of_steel_720p.mkv -c:v:0 libx264 -preset slow -crf 22 -c:a copy tears_of_steel_h264.mkv
-preset sets a encoding speed, more it will be slow better the image quality will be. Are available: ultrafast, superfast, veryfast, faster, fast, medium, slow, slower et veryslow. Medium is the default setting.
-crf set a quality level between 0 and 51 (small number = better quality but more time to process)
A more detailed documentation is available here: https://trac.ffmpeg.org/wiki/Encode/H.264
Encode video in H.265/HEVC
The H.265/HEVC is a new proprietary codec, futur successor of the H264/Mpeg4. The codec that manage it are also new, and all the options and optimisations are not available. To get a better speed of encoding and a better image quality, try to get a newer version of the codec.
avconv -i tears_of_steel_720p.mkv -c:v:0 libx265 -preset slow -crf 22 -c:a copy tears_of_steel_h265.mkv
A more detailed documentation is available here: https://trac.ffmpeg.org/wiki/Encode/H.265 and here http://x265.readthedocs.org/en/default/
Encoding sound
The list of available option for transcoding sound can be found here: https://ffmpeg.org/ffmpeg.html#Audio-Options
The main options:
options |
details |
-acodec ou -c:a |
to choose a codec |
-ar |
set the sample frequency (44100 Hz) |
-ab |
set the bitrate, 64kbps by default |
-ac |
set the number of channels (mono-stereo |
Encode sound in Vorbis
Vorbis is an open source audio codec, with loss, equivalent to MP3/H264:
avconv -i tears_of_steel_720p.mkv -c:v copy -c:a:0 libvorbis -qscale:a 5 -ar 48000 tears_of_steel_vorbis.mkv
- -qscale defines the quality level between 0-10, where 10 is the best quality. By default, qscale is set to 3.
- -ar 48000 allows to resample the audio in 48kHz.
Encode a audio file in MP3
Several codecs can encode the audio to MP3, limp3lame is one of the most used. Here, an example forcing the bitrate to 256Kbits/s, resampling in 44100hz and forcing the production of a stereo channel:
avconv -i 01.ogg -acodec libmp3lame -ar 44100 -ac 2 -ab 256k 01.mp3
Warning metadata can be lost using this method
Still images
Transform a image sequence into a video
avconv -f image2 -pattern_type glob -i "*.tif" -r 24 outputvideo.mkv
In this example, we transcode all the .tif files from the current folder into a outputvideo.mkv. But it's possible to use other format types such as: PGM, PPM, PAM, PGMYUV, JPEG, GIF, PNG, TGA, TIFF, SGI, PTX.
-r 24 allows to specify the framerate
Be careful to the debit, still images can be very heavy. You might need to add a compression behind:
avconv -f image2 -pattern_type glob -i "*.tif" -r 24 -c:v:0 libvpx -crf 10 -vb 4M outputvideo.mkv
External Links
FFmpeg Homepage: https://ffmpeg.org/
FFmpeg Documentation: https://trac.ffmpeg.org/wiki
Libav Homepage: https://libav.org/