MPEG DASH Adaptive Streaming¶
MPEG DASH is MPEG’s standardized Dynamic Adaptive Streaming over HTTP. It is specified in the following international standards:
- ISO/IEC 23009-1 specifies the overall DASH architecture and the XML syntax for the MPD (Media Presentation Document)
- ISO/IEC 23001-7 specifies the Common Encryption for MP4 fragments
- ISO/IEC 14496-12/AMD 3 specifies the extensions to 14496-12 that are necessary to support DASH with fragmented MP4 media
Quick introduction to MPEG DASH¶
At a high level, an MPEG DASH presentation consists of an initial XML manfifest, called the Media Presentation Document (MPD for short), which describes media segments that form a complete presentation. Along with a number of attributes, the MPD allows the MPEG DASH player to compute the URL of each segment, and to download and render them.
mp4dash tool takes care of creating the DASH MPD document, as well as packaging the media segment in their proper form and output location, including pre- and post- processing if necessary (like encryption for DRM support).
mp4dash packager tool¶
mp4dash is the name of the Bento4 tool that can convert one or more input media files into a complete MPEG DASH presentation.
mp4dash for HLS as well
Recent versions of the HLS specification support uusing MP4 fragments. Since those fragments are essentially the same as those used for MPEG DASH, the
mp4dash tool can be used to output HLS presentations as well as DASH, from a single set of input files. Use the
--hls option to enable that mode. See the HLS page for more details.
Running this tool requires that you have python version 3.7 or above installed on your system (python 2.x is no longer supported). Please visit www.python.org if you need to install python.
Python on Windows
On Windows, the launch scripts assume that python3 is launched with the command python, but on all other systems, the command python3 is used.
If you are using the Bento4 precompiled SDK distribution, you will find it convenient to use the
mp4dash command, located in the
bin/ directory, which will automatically run the
mp4-dash.py python script for you. If you are running from a source distribution, you may need to invoke the python script (located under
Source/Python/utils) directly, like this:
python3 <path-to-bento4-python-utils>/mp4-dash.py <arguments>
In the examples below, we assume that you are running from a standard precompiled distribution, thus we will simply use the
Running with self-compiled binaries
mp4dash tool needs to invoke the Bento4 command line binaries
mp4split. If you are invoking the tool as distributed in the pre-compiled SDK, the binaries will automatically be found in the SDK package as installed. If you are running the tool from the Source SDK, with binaries that you compiled from the source, so you’ll need to use the
--exec-dir option to specify the directory where your built binaries reside (platform dependent).
If you just run the
mp4dash tool with no arguments, it will print out the options that are supported.
For a complete list of command line options for
mp4dash, visit the mp4dash command line details page.
Preparing the input files¶
To create DASH MP4 content, you need to start with fragmented MP4 files. The input files you will be working with may or may not already be in fragmented MP4 form. Ideally, your encoder will already produce MP4 files that way.
The command line tool
mp4info can tell you if an MP4 file is fragmented or not: here’s an example of what you would see in the
Movie: part of the
mp4info output for a non-fragmented MP4 file (the line
Movie: duration: 147188 ms time scale: 2997 fragments: no Found 2 Tracks
If you have non-fragmented MP4 files, you can use the mp4fragment tool to fragment them.
mp4fragment video.mp4 video-fragmented.mp4
Dealing with ISMV and ISMA input files¶
ISMV and ISMA files are basically fragmented MP4 files, and as such can be used as input to
mp4dash. However, most legacy tools (encoder and packagers) that produce ISMV and ISMA files generate fragmented MP4 files that lack a
tfdt timestamp box. While these files appear to be normal fragmented MP4 files, the lack of tfdt renders those media files unsuitable for playback with many DASH clients, including most HTML5 based clients.
If you have such files, you can fix this problem by refragmenting your ISMV and ISMA files into compliant fragmented MP4 files, using the
Example: re-fragmenting ISMV/ISMA files
mp4fragment video.ismv video.mp4
This should print something like
NOTICE: file is already fragmented, it will be re-fragmented found regular I-frame interval: 1 frames (at 25.000 frames per second)
You can then use the re-fragmented files as input to mp4dash.
You can tell whether your fragmented input file has a tfdt box or not by using mp4dump. A file with tfdt will show entries like this:
[moof] size=8+816 [mfhd] size=12+4 sequence number = 490 [traf] size=8+792 [tfhd] size=12+4, flags=20000 track ID = 2 [tfdt] size=12+8, version=1 base media decode time = 32281600 [trun] size=12+744, flags=301 sample count = 92 data offset = 832
Generating a DASH presentation¶
Once you have fragmented MP4 files to work with as input, you can generate a DASH presentation, including an MPD and media files or media segments. For single-bitrate streaming, a single MP4 file is required.
The general usage for the
mp4dash tool is:
mpdash <input files>, which produces a DASH presentation in an output directory (
output by default), including the MPD file that points to the media segments (
output/stream.mpd by default).
Example: single MP4 input file
While it is entirely possible to use MPEG DASH for simple, single-bitrate Audio/Video streams, multi-bitrate support is where MPEG DASH shines. The most common use case is to package multiple video streams, each encoded at a different bitrate, into a single presentation, where the player will adaptively switch between segments of the different streams depending on the network and playback conditions.
When supplying multiple video streams as input (one video track per input file) to
mp4dash, it is important that the files have been encoded in a way that is compatible for this use case. You will need a set of MP4 files that have been encoded with closed GOPs (Group Of Pictures) with equal durations. Also, the audio tracks in all the files should be encoded with the same parameters. Once you have your input files, you can use the
mp4dash tool to automatically generate the DASH MPD and, optionally, split the MP4 file into individual file segments.
Example: multi-bitrate set of MP4 files
mp4dash video_1000.mp4 video_2000.mp4 video_3000.mp4
Codecs other than xHE-AAC aren't well suited for seamless dynamic bitrate switching, so when using traditional codecs like AAC or Dolby Digital, it is usually the case that a single audio stream (per codec) will be present as input. If more than one input includes an audio track (which may be expected with multi-bitrate packaging, where each of the input files may contain both an audio and video track), only one of the audio tracks will be retained, unless they differ by more than 10% in bitrate. For the xHE-AAC codec, however, seamless dynamic bitrate switching is possible. For more details on how to support that use case, read the Multi Bitrate Audio page.
The name of input files may be preceded by an input qualifier, which can be used for stream selection as well as setting properties for the input. The general syntax for input qualifiers is:
mp4dash will create an output that contains all the audio tracks found in its each input, as well as one video track from each input (it is assumed that each input file only has one video track, but multi-track input files may still be used, using a selector to select one of the tracks). It is possible to limit the selected tracks by using input stream selectors.
The supported selectors are:
||Select track ID
||Only keep audio tracks with a specific language|
Setting stream properties via input specifiers allows invoking
mp4dash with extra information about the input that may not be otherwise inferred, or should override some of the streams' properties. This can be used, for example, to override the language code for an audio track, or specify labels or group names.
||Set/override the language code for an input|
||Set the language name (i.e human-readable string for the language) for an input|
language=<language-code> selector to select one of the audio tracks. The language code is a 2 or 3 letter code as defined by the DASH specification – (see RFC 5646]. It is possible to combine the language selector with a type selector, by separating the two by a comma (ex:
[type=audio,language=fr]). If the language selector is used by itself without a type selector, the video track as well as the matching audio track will be selected.
Select the French audio track and the video track from video.mp4:
Select the French audio track from
video1.mp4 and the video track from
mp4dash [type=audio,language=fr]video1.mp4 [type=video]video2.mp4
Select track ID 3 from video1.mp4 and all tracks from video2.mp4:
mp4dash [track=3]video1.mp4 video2.mp4
Select the track with
undefined language from
video1.mp4 and remap the language to
mp4dash --language-map=und:fr [language=und]video1.mp4
--verbose option may be useful when debugging language selection options
Encryption and DRM¶
MPEG DASH streams can be encrypted, and played on clients that have a DRM-enabled DASH player. Visit the MPEG DASH Encryption & DRM page for details.
mp4dash packager supports input files with streams encoded with a variety of difference codecs. For details on what's supported and what options may be used, read the Codecs page.
Serving DASH Streams¶
mp4dash produces an output that can be served by any regular HTTP server. In order to make this possible, each one of the media segments and the init segments are stored in separate files. The MPD then refers to relative URLs for those files. Care should be taken to configure the HTTP server to return the correct
Content-Type for the MPD document and segment files. The MPD document (named
stream.mpd by default by the
mp4dash tool) should be served with
Content-Type: application/dash+xml and the init and media segments should be served with a
Serving DASH without splitting segments¶
For players that support the "on demand" profile, input files can be kept as single files on the server, and the players use an index embedded in the files to access individual segments using HTTP range requests.
For the "live" profile, however, there is no such embedded index. With special support from the server, it is possible to serve such DASH streams without splitting the segments into individual files. This is achieved by having a server-side module that can virtualize URLs, so as to be able to expose each segment as an separate URL, mapping it back to a byte range in a source file when requests are made. MPEG does not specify a common standard for such URL virtualization. Each server that supports this model is free do do it any way it chooses. For example, the Microsoft IIS server can be used with a IIS Media Services extension for Smooth Streaming to deliver MPEG DASH streams. More details can be found on the Serving MPEG DASH with Microsoft IIS page. Another useful way to do the same thing is to use the open-source Hippo Media Server package, which knows how to virtualize URLs for MPEG DASH and Smooth Streaming.
--hippo option of
mp4dash is used to generate the server-side manifest needed by the Hippo Media Server.