Windows Media 9 Series Beta: Technical Review

The long awaited release from Microsoft's Windows Media Group, code named Corona, has finally appeared in a beta version. Now dubbed Windows Media 9 Series, it is the first comprehensive platform review in four years. It includes new audio and video codecs, enhanced encoding software and tools, a new server, a new look for the player, and a new SDK for developers.

As expected, this release is much more than a streaming media release. Many of the new features are more broadly targeted at consumer media consumption. These features enable consumers to watch DVDs, listen to downloaded or ripped music tracks, and even include personal video recorder functions built in to the upcoming Windows XP Media Center release. This preliminary review concentrates on the features of most interest to today's streaming media professionals.

New Codecs

On the audio side, the most exciting news is a new low bit rate mixed-mode codec. This codec is designed to encode content that has both voice and music content. Previously, content creators had to decide whether to optimize their low bit rate streams for either voice or music. This codec does justice to both. Even more impressive is that WMA 9 is backwards compatible - no need to upgrade older players or devices.

The question is, how does it stack up against the competition? The table below provides some samples for you to compare. The WMA9 samples were encoded using the new Windows Media 9 Series encoder; the RealSystem, WMA7 and WMA8 samples were encoded by exporting from Sound Forge; and the QuickTime and MPEG-4 samples were exported directly from QuickTime Pro.

Click on any clip to download it for local playback on your machine. The clips are not offered as streaming versions since network conditions add an additional layer of complexity. How these clips perform under streaming conditions is another review altogether.

Bitrate	8kbps voice	16kbps voice	8kbps music	16kbps music
Windows Media 7/8	Click to d/l	Click to d/l	Click to d/l	Click to d/l
Windows Media 9	Click to d/l	Click to d/l	Click to d/l	Click to d/l
RealSystem	Click to d/l	Click to d/l	Click to d/l	Click to d/l
QuickTime	Click to d/l	Click to d/l	Click to d/l	Click to d/l
MPEG 4	n/a	n/a	n/a	Click to d/l

When listening to these samples, listen carefully for the overall frequency range of the samples, and compare the fidelity of each. Voice artifacts are audible as a "lisping" or "hollow" quality, pre-echo, and audible ringing. Music artifacts are audible as distortion, rumble, as well as pre-echo and ringing.

The WMA 9 hybrid codec is apparently two separate codecs, with intelligence to dynamically switch between them based on the source content. It is definitely a huge improvement over WMA 8. For those of you working on the Windows Media platform, you have a great new option for low bit rate mixed-mode content. When compared to the other platforms, however, the competition gets a little tougher.

Codec designers have to trade off audio fidelity with frequency range, and the subtle differences between the platforms reflect this. Some codecs sound a bit brighter but perhaps with more artifacts; others a bit less bright but less audible artifacts. Personally, I prefer fidelity to extended frequency range, but have been astounded repeatedly at consumer preference for extended frequency ranges. Then again, consumers prefer "loud" speakers to accurate speakers - ask any stereo salesperson.

Bearing this in mind, both the WMA9 hybrid voice and RealAudio 8 music codecs do a good job, with the QuickTime and MPEG-4 codecs bringing up the rear (The MPEG-4 codec available in the QuickTime encoder does not offer any bit rates below 16kbps). The new Windows Media hybrid voice codec does an outstanding job on the voice, but the strings in this example have an annoying ring. RealAudio 8 music sounds a bit hollow on the voice, but the strings sound more natural. Between Windows Media and RealAudio, there is no clear winner - it comes down to individual preference.

In addition to this low bit rate hybrid codec, Windows Media 9 Series adds a host of other audio codecs, targeted to higher-end non-streaming applications. These are called Windows Media 9 Professional codecs, and include a new lossless codec for audiophiles, a surround codec that can handle up to 8 channels, and support for 96KHz/24 bit audio. Additionally, the WMA 9 codecs now support 2-pass and variable bit rate encoding for higher fidelity.

New Codecs - Video

The clips were sourced as full screen NTSC clips in AVI format with photo-JPEG compression on individual frames. To bypass any processing and resizing done in the encoders, the clip was exported from Vegas Video as an uncompressed 320x240 AVI. All the clips were encoded at constant bit rates in one pass, since QuickTime Pro does not offer 2-pass encoding. No additional processing was done during the encoding phase, and all default settings were used except for video bit rates, which I tried to match as best as possible.

The montage has a number of things that are difficult to encode: flames, explosions, moving fog, rapidly changing computer-generated graphics, hand-held footage, and a long tracking shot. When watching these videos, look out for blocking artifacts, smearing, and frame rates. Again, click on any clip to download it for local playback. As per the audio samples, no streaming versions are offered.

Bitrate	34kbps	100kbps	225kbps	350kbps
Windows Media Video 8	Click to d/l	Click to d/l	Click to d/l	Click to d/l
Windows Media 9 Series	Click to d/l	Click to d/l	Click to d/l	Click to d/l
RealVideo 9	Click to d/l	Click to d/l	Click to d/l	Click to d/l
QuickTime	Click to d/l	Click to d/l	Click to d/l	Click to d/l
QuickTime MPEG 4 (using the "improved" profile)	Click to d/l	Click to d/l	Click to d/l	Click to d/l

At 34kbps, encoding this clip at 320x240 is asking for trouble. However, codecs are getting to the point where it's not out of the realm of possibility. The Windows Media 9 Series clips show marked improvement over the Windows Media 8 clips. The WMV8 clips are blockier, whereas the WMV9 clips appear smoother. RV9 at this bit rate exhibits less blockiness than WMV9 but more smear. Sorenson V3 can't really compete at this bit rate, unfortunately. The QuickTime MPEG-4 clip does a great job of keeping up with the frame rate, but exhibits a bit rate of 9.1Kbytes, or 72kbps, even though it promised a bit rate of 32kbps when it was exported from QuickTime Pro.

At 100kbps, things start to get interesting. The WMV9 clip again shows improvement over the WMV8 clip. There is more detail, and the frame rate is slightly better. Sorenson starts to look good, but still exhibits many more artifacts than any of the other codecs. The MPEG-4 stream is worth watching, but even though it outperforms the Sorenson clip it is still outclassed by the Real and WMV streams. As for Real versus WMV9, it's a photo finish, possibly with WMV9 winning by a nose.

At broadband rates, the story is pretty much the same. WMV9 and RV9 lead the pack, though both Sorenson and MPEG-4 get more competitive at broadband rates. At 225kbps, both Real and WMV9 can maintain a solid 29.97 frame rate, but both still exhibit extensive artifacts. At 350kbps, we're nearly there - but then again, this is an extremely tough clip to encode.

Please note that no optimization was done during the encoding phase. No doubt better results could be obtained with different tools and a bit more time invested. In particular, little can be done without the professional version of Sorenson V3. Your mileage will vary - the idea behind this encoding matrix is to compare bare-bones default encoding settings.

Final opinion? WMV 9 is an excellent video codec, and seems to have the lead in quality, though RV9 puts up a great fight. I qualify this statement only because I have yet to subject these new codecs to real-world streaming conditions, and the advantage it has over RV9 is slim.

The Windows Media codec group has done a great job improving their technology, and to be fair they've done more at the high end than at the lower end of the bit rate spectrum, where today's streaming media lies. There are some interesting developments in the higher bit rate range, such as support for interlaced video, non-square pixels, and high definition video (1920 x 1080 progressive scan). However, these are not currently within the purview of today's streaming media professional.

Other Improvements

Scalable video support, however, is very exciting. While it has been possible for some time to provide different streams to users with different connection speeds, the solution has never been optimal. RealNetworks SureStream technology provides some slick stream-switching capabilities when throughput is compromised, but you have to author all your streams at the same resolution. QuickTime allows you to specify different streams for users based on any number of qualifiers, but the files themselves must be encoded separately. The new Windows Media 9 Series scalable video allows you to specify different video resolutions and audio codecs for different connect speeds, and places them all in one convenient file. Not only that, but you can add or subtract streams from a scalable video file.

What this means is that you can encode a single file that contains a 176x132 version for dial-up users, a 320x240 version for broadband users, and a full screen (640x480) version for LAN users, each with different audio codecs, all contained within a single file. This is the slickest implementation of multiple bit rate files yet.

The new still image codec allows you to create slide show presentations using high quality stills. At the reviewer's conference, a demo was shown whereby images can be zoomed into, panned across, and a slideshow easily created. Backing tracks and narration can be added, and transitions between slides can be specified. The functionality demonstrated was very similar to RealNetworks RealSlideShow. Not exactly ground breaking, but a welcome addition to the Windows Media family of codecs nonetheless.

In addition to the above mentioned new technologies, the screen codec has been improved, which can now do lossless screen encoding at 96kbps, and encode full screen captures at less than 28kbps. New Encoding Software and Tools The Windows Media encoder UI has changed somewhat, going for a slightly cleaner look. A handy toolbar has appeared at the top, where the start, stop, and pause encoding buttons now reside. Another nice feature is the properties icon on the toolbar, which puts all the session settings within easy reach - no more pop-up windows necessary.

There are other new features aimed squarely at the streaming media professional, such as timecode, IEEE 1394, and Edit Decision List (EDL) support. Not only can timecode be stamped into Windows Media files, but the Windows Media Encoder can control devices via FireWire, which means you can choose segments of a videotape for the encoder to automatically edit together and encode for you. How cool is that?

In addition to FireWire support, there is also VTR support, meaning you can control legacy devices such as beta decks via RS 422. The Windows Media Encoder now also supports plug-in processing. This means you'll be able to use some of your favorite audio and video processing algorithms directly inside the encoder. Plug-in architectures are standard in audio and video editing platforms - it's about time one of the streaming media platforms added this functionality.

The new release also includes utilities that make it simple to edit encoding profiles, add or subtract streams to scalable video files, and edit Windows Media streams. They are all simple, no frills tools that will be very useful to streaming media professionals. There is also tighter integration with the new Windows Media Server. For instance, you can now "push" streams to the Windows Media Server, and automatically create publishing points on the server instead of having to set them up manually. In addition, you can add DRM protection to live streams - previously not possible.

New Server

Fast Start seeks to minimize the buffering time before a streaming media file plays. The player must always buffer 3-5 seconds due to codec and packetization issues, but with Fast Start the data is sent much faster than the stream's native bit rate. For instance, if a user is watching a 20kbps stream on a broadband connection, the data can be sent at 100 - 200kbps, which means the first 5 seconds can be buffered in less than one second.

In addition to minimizing the initial buffering time, data can be sent over and above the streaming data rate and cached to offer more protection from bandwidth shortages. This is known as Fast Caching. The player uses Internet Explorer's cache, so any restrictions imposed by IE are inherited by the player.

Most interesting for live streaming media producers is Fast Reconnect, which attempts to reconnect any lost net connections. Using Fast Reconnect, a player that loses its connection to a server plays bits from its (Fast) cache while attempting to reconnect to the server. Fast Reconnect technology also works between the encoder and server. Obviously, in a live situation you can't send bits faster than they're being encoded, but having an encoder that proactively tries to re-establish a lost connection to the server is something that should have been available years ago. The demo shown at the conference was very impressive - I know I can't wait to test this functionality out in the field.

One thing I must point out is that most of these new technologies, as well as RealNetworks TurboPlay, are based on the availability of additional bandwidth. This is not always the case for Internet streaming. Additionally, most streams are authored to make the most of the user's connection. If a user connects at a broadband rate, chances are that they will be steered towards a broadband stream, leaving little or no extra bandwidth. In cases like these, these technologies may not make a noticeable difference. These technologies can be very attractive, however, when used on an Intranet, where there is generally excess bandwidth available.

The other two major announcements are server-side play lists and improved advertising support. The new server-side play lists are created through the server interface, and stored in a SMIL 2.0 file. They can be changed dynamically, personalized for individual users, and feature both live and on-demand content. Improved advertising support allows dynamic insertion and deletion, as well as cookie support. Previously ads were inserted via the ASX file; inserting them from the server side is more efficient and flexible.

Other miscellaneous server improvements have been made, such as improved packet recovery, better multiple bit rate (MBR) stream support that now allows multiple audio streams, and better bandwidth detection. One surprise announcement is support for the RTSP protocol. For the time being, the Windows Media Group is still recommending the MMS protocol for its rollover capabilities, but eventually MMS will be phased out.

Administering the Windows Media Services 9 can now be done via the traditional Microsoft management Console or via a web-based HTTP interface, which allows remote access across firewalls. Finally, this release includes code optimizations that allow greater numbers of streams to be served per server.

Early Impressions

Interestingly enough, RealNetworks announced immediate support for Windows Media 9 Series. While there are a number of features that will not be supported, such as the Fast Streaming technologies, the fact that interoperability is becoming the norm instead of the exception is promising. Add to this the progress MPEG-4 is making as a format, and the streaming media landscape is looking better and better. We may actually be approaching the time when the platform wars are less important than the content we're trying to deliver. Let's hope so.