iPhone, iPad, iPod – HTTP Live Streaming (HLS) with free tools on Windows

18 Comments on “iPhone, iPad, iPod – HTTP Live Streaming (HLS) with free tools on Windows”

1. What resolution/frame rate is your source? If it’s anything but something small (say, 320×240 or less), encoding _any_ video but the lowest resolution at 96kbps is difficult as you likely reach the limit of the codec’s compression. For any given clip, if you want to see the limit of compression with x264, set the quantiser to 51 and use that as the rate control method. Whatever output bit rate you observe is the maximum compression that x264 can achieve. Note that slower presets will improve compression at the cost of encoding speed.

   Also note that in FFmpeg, all bit rate values are in bits per second. You can express 96kbps as either ‘96000’ or ’96k’.

   The FFmpeg AAC encoder is known to be bad quality in the past releases. It is getting better but it’s still not good enough really and needs significant further development.

   If you have specific questions about x264, you can get support on IRC in #x264 on irc.freenode.net.

   Reply

   • @Rob:
     Thanks for the pointers.

     I *could* get FFMPEG to come close to the requested bitrate by fiddling with the quantizer (qmin to be exact). And yes, with the quantizer, I was able to reach my target bitrate or lower…. so it appears that my video dimensions may not be a factor. Changes in movement within the video will increase bitrate, and unless I misunderstand, a quantizer setting isn’t going to take that into account.

     In summary, FFMPEG doesn’t appear to automatically adjust quality based on a bitrate target, on the fly, as good as using x264 directly. FFMPEG seemed to take a hint from the bitrate, as it did make some difference in the file size, but never really honored the bitrate as well as the x264 executable. I tried many different things, including two pass encoding (which I believe your encoding guide recommends for constant bitrate), as well as the ioncannon site mentioned above. I mostly specified bitrate as “-b 96k” as an example.

     Reply

2. Thought I’d add this as a comment to help others figure a few things out. (I used your commands above as a starting point).

   The above is great, except for one thing. Lots of intermediate files, lots of seperate commands.

   For the Audio Encode, instead of 2 steps.

   ffmpeg.exe -i INPUT.AVI -vn -f wav -acodec pcm_s16le – | neroAacEnc.exe -if – -cbr 64000 -he -ignorelength -of AUDIO.AAC

   Just a single step using Piped output from FFMPEG to NeroAAC
   (-ignorelength is used in the AAC encoder) – An alternative to Nero, is QAAC, which uses the quicktime encoder installed if you have quicktime installed. (if you have iTunes, you have this) https://sites.google.com/site/qaacpage/
   Command line parameters will be different.

   For the video,
   I personally needed to do things like skip seconds and frames and I wanted to do it consistently. (so same command to skip seconds for the audio and for the video)
   So I did piped output for x264 from ffmpeg. Again based off your command above.

   ffmpeg.exe -i INPUT.AVI -an -f yuv4mpegpipe – | x264.exe – –stdin y4m –level 30 –profile baseline –bitrate 96 –keyint 30 -o OUTPUT.MP4

   The required magic in the above:
   -f yuv4mpegpipe means you don’t need *much* extra in the x264 command (raw doesn’t include things like resolution etc, the pipe above does I think)
   for x264, –stdin y4m ensures it gets the right details in from the PIPE input.

   For people new to piped methods:
   At the end of ffmpeg I use “-” for the output file name,
   This tells it to pass the output to the next program.
   the “|” represents the next program.
   and in x264 instead of specifying the input file, I use “-” again to say: “Your input file comes from the previous program”.

   So don’t delete any random -‘s or the | from the above, or it won’t work.

   – Pete

   Reply

   • More fun with pipes?

     2 commands. Audio, then video. You can run the video command as many times as you want.
     Make sure you change the segmenter output filenames!

     This means you only end up with the segmented TS files.
     One issue I had was the -vbsf mpeg4toannexb Had to drop that from the video encode command.
     (I haven’t tested the final pipe to the segmenter in this configuration, but it outputs the unsegmented TS files for me fine otherwise – and my past experience with the segmenter has worked with a pipe)

     Create the audio:

     ffmpeg.exe -i INPUT.AVI -vn -f wav -acodec pcm_s16le – | neroAacEnc.exe -if – -cbr 64000 -he -ignorelength -of AUDIO.AAC

     Then use the Audio, mux it with the video, output it all to the segmenter in segmented files. No intermediate files except the audio (which you needed in the first place for the lowest stream!)

     ffmpeg.exe -i INPUT.AVI -an -f yuv4mpegpipe –
     | x264.exe – –stdin y4m –level 30 –profile baseline –bitrate 96 –keyint 30 -o –
     | ffmpeg.exe -y -i – -i AUDIO.AAC -f mpegts -vcodec copy -acodec copy –
     | segmenter.exe – 10 STREAMNAME/LOW STREAMNAME/LOW-TMP.M3U8 http://mydomain.com/virtual/path/to/STREAMNAME/

     if you are really creative you can even use the “echo” command in a batch file and stdout redirection to create the overarching m3u8 file (the one that joins the different bitrate streams)

     You can automate a lot of the process using dos…

     Reply

3. This is an excellent article! I followed and it works. My videos are on-demand. I do not see end time in video player in iPhone and it shows as LIVE!

   Please help!

   Reply

   • Try running it through Apple’s mediastreamvalidator. It should tell you what is wrong.

     http://developer.apple.com/library/ios/#technotes/tn2224/_index.html

     You can also try to point the QT player on iOS to the sub-streams… IE instead of the variant playlist m3u8, point it at one of the generated m3u8’s. If it still shows no end time, check the contents of the generated m3u8. It’s a simple text file, and if properly generated, you should be able to calculate the time (to the nearest second or so) yourself by totaling up the chunk durations. The total duration in each of the m3u8’s referred to in your variant playlist should be identical.

     Gobs of information on the Apple website and their developer forums. Here’s another link:

     http://developer.apple.com/library/ios/#documentation/NetworkingInternet/Conceptual/StreamingMediaGuide/Introduction/Introduction.html

     Good luck!

     Reply

4. Finally We figured! The end tag was missing from M3U8 file which made Apple player think it is LIVE!

   Thank you for your support.

   BTW, do you know how to set screen size (iphone: 480 by 320) so the HLS is set for iPhone and for iPad it is 680 by 320.

   Reply

   • Check the -s option in FFMPEG’s documentation: http://ffmpeg.org/ffmpeg-doc.html

     Reply

5. This creates static TS and m3u8 files for Apple HLS. If I want to streamline the work from encoder to a storage, say, ingest the ts/m3u8 while they are being generated by the pipeline process, to an internet storage. Is there anyway to do it now?

   Reply

   • I believe FFMPEG can read from a steam, and yes, there is a way to get these to work using pipes. But you may end up relying on FFMPEG’s encoding, rather than using x264 / NeroAAC… FFMPEG will definitely work, but it doesn’t do a great job giving you the bitrate you ask for. See above comments from PeterA on using pipes.

     I’m pretty sure Wowza 3, which just came out, supports this out of the box, but it isn’t free.

     Good luck!

     Reply

6. step number 10:
   echo #EXTM3U> STREAMNAME\STREAMNAME.M3U8
   echo #EXT-X-STREAM-INF:PROGRAM-ID=1, BANDWIDTH=320000>> STREAMNAME\STREAMNAME.M3U8
   echo med.m3u8>> STREAMNAME\STREAMNAME.M3U8
   echo #EXT-X-STREAM-INF:PROGRAM-ID=1, BANDWIDTH=864000>> STREAMNAME\STREAMNAME.M3U8
   echo high.m3u8>> STREAMNAME\STREAMNAME.M3U8
   echo #EXT-X-STREAM-INF:PROGRAM-ID=1, BANDWIDTH=160000>> STREAMNAME\STREAMNAME.M3U8
   echo low.m3u8>> STREAMNAME\STREAMNAME.M3U8
   echo #EXT-X-STREAM-INF:PROGRAM-ID=1, BANDWIDTH=64000>> STREAMNAME\STREAMNAME.M3U8
   echo audio.m3u8>> STREAMNAME\STREAMNAME.M3U8
   How i can apply it?
   ‘#EXTM3U’ is not recognized as an internal or external command,
   operable program or batch file.

   Reply

   • I’m not sure what you are asking. echo is the command, not #EXTM3U. This only works on a Windows command prompt.

     Reply

     • what is #EXTM3U> STREAMNAME\STREAMNAME.M3U8 means?

       Reply

       • I recommend reading up on the windows command line for more information… Please read:

         http://www.bleepingcomputer.com/tutorials/windows-command-prompt-introduction/

         and for details on the redirect operators:

         http://technet.microsoft.com/en-us/library/bb490982.aspx

         and the echo command:

         http://technet.microsoft.com/en-us/library/bb490897.aspx

         For more information on the m3u8 file, please check the apple website:

         http://developer.apple.com/library/ios/#technotes/tn2224/_index.html

         Good luck!

         Reply

       • Actually i have done all of those steps but im still not able to stream anything from my iphone

         Reply

7. Anyone have a lead on an automated process for this? While doable using the command line, I have loads of video that will need to be processed, and would greatly like to save some time.

   Reply

   • Here’s an example batch I found to do that.

     http://forum.videohelp.com/threads/341663-How-to-Use-x264-Command-Line?p=2127375&viewfull=1#post2127375

     PeterA: thank you for that helpful explanation of piping. That’s helped my setup.

     Andrew: qtaacenc is an alternative aac encoder, which exploits an install of Quicktime/iTunes (which are *free*), to encode to aac.

     Here’s a test command that ripped audio out of an .mp4 as PCM data, then piped that data to qtaacenc, and encoded it to aac (in .m4a format):

     ffmpeg -i test.mp4 -f wav – | qtaacenc –tvbr –highest –samplerate keep – test.m4a

     Reply

8. Thank you very much for this post, which helped me a lot.

   I don’t know whether you ran across it, but qtaacenc.exe, which is a command-line encoder, is a workable option. It relies on an installation of QuickTime and/or iTunes, with Apple Application Support. I ran across this option, with a user’s opinion that Apple’s aac encoding is equal or superior to Nero, here:

   http://ubuntuforums.org/showpost.php?p=12228107&postcount=6

   I got piped output from a .mp4 file, via ffmpeg, to encode with the following command: (adapting PeterA’s approach–by the way, PeterA, I’ve run across many explanations of piping, which have all always been difficult to understand, and yours is the *only* clear explanation I’ve ever read about it–thank you!)

   ffmpeg -i ep2c1Logo.mp4 -f wav – | qtaacenc – test.m4a

   I assume that could be further adapted via mp4box etc. to mux the m4a stream (either through an intermediate file, or through another pipe?) with a video stream into an .mp4 container.

   I don’t know the licensing terms for Apple’s aac encoder, but I hope it’s more liberal than Nero’s. The latter’s license agreement says you may not use Nero aac for commercial purposes. I suppose I’ll look up the Apple aac encoder’s license terms . . .

   Reply