At WWDC 2016, Apple announced support for fragmented MP4 (fMP4) as an alternative to MPEG-TS, which prior to their announcement was the only supported format.
So why use fragmented MP4 files? Well, according to Apple’s video encoding requirements in their HLS Authoring Specification, if you want to use HEVC/H.265, you have to use it (1.5). Fragmented MP4 files are also compatible with MPEG-DASH – an alternative to HLS – so you can use the same files; only the manifest file (playlist) is different. This means less encoding and less storage requirements, which should reduce costs.
In this post, I’ll demonstrate how to generate fMP4 files using ffmpeg. You’ll need a fairly recent version of ffmpeg that supports fMP4 – I used version 4.1.1 to create the files in this post.
I often get email from people asking if there any examples in the book about how to play HLS video in the browser and my answer is always the same: no. In this short post, I’ll try to explain why not with an example of how to play encrypted HLS video in the browser, which is also something I get asked about a lot.
The main reason there aren’t any specific examples in the book that show how to play HLS video in the browser is because most of the time, all you need to do is tell the player where the playlist is and hit play; it takes care of the rest. Plus, there are a lot of players out there that support HLS video.
Most of the effort involved in playing HLS video goes into encoding the videos, segmenting them, preparing the playlist, and so on, not configuring the front-end. And this is what the book focuses on.
Let me highlight this with an example. I’m going to use Video.js, but it should work in just about every other HLS capable player. To play HLS video using Video.js you’ll need to use the Video.js HLS source handler. I’ve already written something about how to encrypt video for HLS so I won’t go over the details here – I did say this was going to be a short post.
As you can see (highlighted) all you have to do to play the video is just tell the player where the playlist is. Even though this particular video is encrypted, there is nothing more you have to do on the front-end.
In this instance, the playlist and the encrypted video are stored in an S3 bucket. (If you are interested, I wrote a post recently about how to serve HLS video from an S3 bucket that goes into more detail.)
This post will describe how to configure an S3 bucket to serve video using HLS.
Amazon S3 is a storage solution that allows you to store large amounts of data for relatively little cost, perfect for something like video files, which tend to be quite large. Files (or objects as they are often referred to) are accessed over HTTP, which makes it a great solution for storing (and serving) your HLS videos.
You’ll need an Amazon Web Services (AWS) account to use S3. If you haven’t got one, you can sign up here. Sign in to your account and navigate to the S3 console. Create a bucket. Next, upload your video segments and playlist etc. to the bucket.
Make sure the content type of the playlist and the video segments (.ts) is set to “application/x-mpegURL” and “video/MP2T” respectively. You can do this by selecting the Properties tab for each file and then clicking on Metadata.
Before you can start serving your videos, you need to grant read access to the files in the bucket; files are private by default. Select the Permissions tab and set the bucket policy to the following:
This presents a bit of a problem. If you request a resource (a file) from a different domain, you need permission to do so; it violates the same-origin policy that browsers enforce. Cross-origin resource sharing (CORS) is a mechanism that allows user agents to request permission to access resources that reside on a different server. In this instance, you need to grant permission to the player to allow it to access to the video(s) in the S3 bucket.
Thankfully, Amazon S3 supports CORS so you can selectively allow cross-origin access to the files in your S3 bucket. To enable CORS on your bucket, select the Permissons tab then click on CORS configuration. By default, the configuration file allows access from anywhere. If I wanted to restrict access to the videos to this domain, I would modify the CORS configuration to look like this:
Once you have granted public access to files in the S3 bucket and configured CORS appropriately, you should now be able to serve your HLS videos from an S3 bucket.
One More Thing
HLS playlists support the EXT-X-BYTERANGE tag. This indicates to the player that each video segment is part of a (larger) resource. (There’s more about this in the book if you are interested.) Here’s an example from a playlist that uses the tag:
With the current configuration, any requests for the video file (main.ts) will result in a 403 Access Denied response from S3. To grant access to the file, you need to allow the use of the HTTP Range header. Modify the CORS configuration file so it looks like this:
Now your video player will be able to make range requests for parts of the video file.
Finally, here’s an example of playing an HLS video from an S3 bucket:
Amazon S3 is a cost effective solution for storing videos as video files can take up a lot of space. Because access to files in S3 is over HTTP and with support for CORS, it also makes it a viable solution for using it to serve your HLS videos to your viewers.
S3 also has other features that I haven’t mentioned here, such as versioning and archiving, that could be useful for managing your video assets. Check out the S3 documentation for more information.
In this post, we’ll look at what encryption HLS supports and how to encrypt your videos with ffmpeg.
Encryption is the process of encoding information in such a way that only authorised parties can read it. The encryption process requires some kind of secret (key) together with an encryption algorithm.
There are many different types of encryption algorithms but HLS only supports AES-128. The Advanced Encryption Standard (AES) is an example of a block cipher, which encrypts (and decrypts) data in fixed-size blocks. It’s a symmetric key algorithm, which means that the key that is used to encrypt data is also used to decrypt it. AES-128 uses a key length of 128 bits (16 bytes).
HLS uses AES in cipher block chaining (CBC) mode. This means each block is encrypted using the cipher text of the preceding block, but this gives us a problem: how do we encrypt the first block? There is no block before it! To get around this problem we use what is known as an initialisation vector (IV). In this instance, it’s a 16-byte random value that is used to intialize the encryption process. It doesn’t need to be kept secret for the encryption to be secure.
Before we can encrypt our videos, we need an encryption key. I’m going to use OpenSSL to create the key, which we can do like so:
$ openssl rand 16 > enc.key
This instructs OpenSSL to generate a random 16-byte value, which corresponds to the key length (128 bits).
The next step is to generate an IV. This step is optional. (If no value is provided, the segment sequence number will be used instead.)
Make a note of the output as you’ll need it shortly.
To encrypt the video we need to tell ffmpeg what encryption key to use, the URI of the key, and so on. We do this with -hls_key_info_file option passing it the location of a key info file. The file must be in the following format:
Path to key file
The first line specifies the URI of the key, which will be written to the playlist. The second line is the path to the file containing the encryption key, and the (optional) third line contains the initialisation vector. Here’s an example (enc.keyinfo):
Note the URI of the encryption key. The player will retrieve the key from this location to decrypt the media segments. To protect the key from eavesdroppers it should be served over HTTPS. You may also want to implement some of authentication mechanism to restrict who has access to the key. If you’re interested, the book goes into some detail about how to achieve this. Click here to buy a copy.
To verify that the segments really are encrypted, try playing them using a media player like QuickTime or VLC. You shouldn’t be able to. Now run the command above without the encryption and then try playing a segment. Notice the difference.
Even though HLS supports encryption, which provides some sort of content protection, it isn’t a full DRM solution. If that kind of thing interests you then you may want to take a look at Apple’s FairPlay Streaming solution.
In a previous post I showed how to segment video for HLS using ffmpeg’s segment muxer. In this one, I’ll demonstrate how to use ffmpeg’s hls muxer. It has more features specific to HLS like support for encryption, subtitles, specifying the type of playlist, and so on.
To follow along, you’ll need a recent version of ffmpeg with support for HLS. (I used version 3.1.1 on Ubuntu 14.04.) To see if your version supports HLS, run the command below. You should see something like this:
$ ffmpeg -formats | grep hls
E hls Apple HTTP Live Streaming
D hls,applehttp Apple HTTP Live Streaming
If ffmpeg complains about an unrecognized option when executing the commands in this post, you can see what options are supported by running the following from a terminal:
Once you’ve deployed your videos, it’s always a good idea to check them for any errors and make sure they are valid. This is especially true if you are developing an iOS application that streams video as Apple can reject your app if there are any problems with your streams. For example, if your application streams video over a cellular network, you must provide a stream with a maximum bit rate of 192 kbps. If you don’t, you aren’t going to make it into the App Store.
We can use Apple’s mediastreamvalidator tool to validate our video streams. It checks that the playlist and the media segments conform to the HTTP Live Streaming specification and will report any problems it finds so we can fix them.
To check if a video stream is valid, call mediastreamvalidator with the URL of the playlist. If you run the following command, you should see something like this:
$ mediastreamvalidator http://hlsbook.net/wp-content/examples/sintel/sintel_index.m3u8
[...] Started root playlist download
[...] Started media playlist download
[...] All media files delivered and have end tag, stopping
Processed 7 out of 7 segments
Average segment duration: 10.000000
Total segment bitrates (all discontinuties): average: 867.51 kb/s, max: 1199.59 kb/s
Discontinuity: sequence: 0, parsed segment count: 7 of 7, duration: 70.000 sec, average: 867.51 kb/s, max: 1199.59 kb/s
Track ID: 1
Video Codec: avc1
Video profile: High
Video level: 3.1
Video resolution: 1024x436
Video average IDR interval: 3.646930, Standard deviation: 3.265633
Video frame rate: 24.000
Track ID: 2
Audio Codec: AAC-LC
Audio sample rate: 48000 Hz
Audio channel layout: Stereo (L R)
No problems to report here. In addition to validating the stream, it also reports information such as the average bit rate, segment duration, and resolution. You can use this information to also check that you’ve encoded your videos correctly.
The latest version (1.1) of mediastreamvalidator supports outputting the data to a file in JSON format. You can then pass this JSON file to another of Apple’s tools hlsreport, which will generate a nicely formatted summary for you in an HTML page. Let’s look at the commands we need to run to do this:
This time we run mediastreamvalidator with the -O option specifying the name of the file we want to write the data to. (I’ve omitted the output.) Next, we run hlsreport.py on the data to generate the HTML page. Here’s an example report showing the findings of the validation tool.
(Interestingly the report flags a number of issues that weren’t reported when only running the validation tool. This could be because it appears to be checking the results against the HLS Authoring Specification for Apple TV. Whatever the reason, probably best to run both just to be absolutely certain.)
So now you know how to validate your HLS video streams, you should have no problems getting your app into the App Store.
In this post we’ll look at how to add session data to a playlist and then how to access it from a client application. We can use this feature to add arbitrary metadata to the playlist, such as the title of the movie, who the director was, and so on. To add session data to a playlist, we can use the EXT-X-SESSION-DATA tag.
Let’s say we have our movie and we want to add the title and the name of the director as session data to the playlist. In this example, we’ll use the computer animated film, Sintel. Let’s start by embedding the values directly in the playlist:
The DATA-ID attribute is the key and the VALUE contains, surprisingly, the value. The convention is to use reverse DNS notation for the name of the DATA-ID attribute. In this instance, we’re using net.hlsbook.movie.title to represent the title of the movie and net.hlsbook.movie.director for the name of the director. (On the client – as you’ll see shortly – we’ll use the name of the DATA-ID attribute to look up its value.) Continue reading →
Quick tip. How do you start playing a video from a specific point in time with HLS? Use the EXT-X-START tag. It’s optional, but if it’s present in a playlist it specifies the preferred point in the video to start playback.
The following attributes are defined in the spec for the EXT-X-START tag:
TIME-OFFSET – (required) The offset in seconds to start playing the video. If the value is negative, specifies the time from the end of the playlist. The offset must not be larger than the playlist duration.
PRECISE – (optional) Valid string values: YES, NO. If YES, playback starts at the point specified by TIME-OFFSET but does not render frames in the segment prior to the offset. Default is NO.
Let’s look at an example. Assume that we want to start playback precisely 25 seconds from the beginning of the video. You would add the following tag to the playlist:
If we want to start 15 seconds from the end of the playlist and we aren’t concerned with precision, the tag will look like this:
To see how this works in practice, let’s look at some examples. (I recommend watching the videos with Safari on an iOS device. I had mixed results with Safari on the desktop.) The first example starts playing at the beginning of the video, the second one – same video, different playlist – starts playing 25 seconds in.
There’s one more thing to consider if you want to use the EXT-X-START tag: It first appeared in version 6 of the protocol so you need to set the EXT-X-VERSION tag in the playlist accordingly.
Here’s the complete playlist for the example with the EXT-X-START tag:
If you want to play the videos in a browser other than Safari, you may want to check out an earlier post that describes how to do this. If it doesn’t work as expected, this may be because the client doesn’t support the appropriate version of HLS. Your mileage may vary.
I’ve just watched a tech talk session about media playback on Apple TV.
There were some interesting uses of HLS. The video player has a sliding thumbnail view, which uses I-frame playlists, that allows viewers to seek to a specific point in the video, and there are some additional properties for displaying metadata about the video, such as chapter markers and interstitial content, but what was particularly interesting was the integration with Siri. In the presentation, the presenter pauses the video and asks Siri what the character in the video just said. This caused the video to seek backwards for 15 seconds and enable the subtitles (temporarily) for that part of the video. After 15 seconds, the subtitles were turned off. A novel use of subtitles!
It’s given me a few ideas for some new content for the book. (Looks like I’ll have to get myself an Apple TV now!)
Check it out if you have the time. It’s worth watching.
HLS video works in all browsers on iOS but on the desktop, the only browser that supports HLS video natively is Safari. The good news is that it is possible to play HLS video in other browsers, such as Chrome or Firefox, but you need to use a workaround. There are two options available:
Flash player with a plugin that handles HLS video
Use the HTML5 video element and Media Source Extensions (MSE)
We’ll look at an example of each in turn and discuss how they work in more detail.
Flash Player with Plugin
Flash works across practically all desktop browsers (and some mobile ones) so you can be confident that your HLS videos will be viewable across a variety of different browsers and platforms (Windows, Mac, Linux).
For the example, I’m going to use the Flashls plugin. It works with a number of different Flash players and supports a variety of HLS features such as adaptive streaming, encryption, and so on. You can view the example here.
Given that Flash is being phased out, this solution may not be the best one going forward. Having said that, it’s unlikely that Flash is going to disappear completely in the next year or two but it is on the way out. The future is HTML5 video and that’s what we’ll look at next.
HTML5 + Media Source Extensions
To demonstrate how to use the HTML5 video element to play HLS video, I’m going to use the hls.js library. Check out this introduction for more information.
The drawback of this approach is browser compatibility. Although the MSE specification provides some guidelines on how the API should behave it does vary across browsers. You also need a relatively recent version of a browser both on the desktop and mobile device. (On iOS, MSE is not supported at all!)