Streaming Media - October 2008 - (Page 74)

Finding the Lowest Possible Data Rate The following is an excerpt from Critical Skills for Streaming Producers, a mixed-media tutorial resource available at www.streamingmedia.com. As we’ve seen, encoding at too high a data rate not only costs extra bandwidth but also can slow the display rate on lower-power computers and increase the processor load required to play the file on all computers. There’s another factor you should consider as well. Figure 1. Data rate and viewer satisfaction Specifically, have a look at the slide shown in Figure 1. The slide is from a workshop given by Hendrik Knoche, a graduate student at University College London, who has published a host of articles on streaming quality that you can see at www.cs.ucl.ac.uk/staff/H.Knoche/publications.htm. Briefly, the slide assesses viewer satisfaction relating to news content containing legible text on mobile TV as displayed on a number of devices. The left axis is video quality acceptability while the bottom axis is the encoding bitrate. As you can see, viewer satisfaction increases sharply at all viewing resolutions as the data rate jumps from 0 to 100Kbps, but then the rate slows dramatically, especially for the 320x240 video displayed on the HP iPAQ, where tripling the data rate increases quality acceptability by about 10%. For the 240x180 and 208x156 videos, viewer satisfaction actually decreased with additional data rate. In financial terms, your initial investments in bandwidth enjoy a high rate of return, but once quality is “acceptable,” the rate of return drops and can even go negative. How do you find the tipping point where quality is good enough? In their article “Can Small Be Beautiful,” Knoche and his co-authors asked their participants to specify what made the video they were watching unacceptable. With news videos, almost 100 of 128 viewers commented about text detail, which, according to the article, included the news ticker, headlines, clock, logos, and captions. Of relevance to this discussion, facial and general detail drew about 30 comments each, as did audio fidelity. This tells me two things. First, when planning your shoots, eliminate as much detail as possible, since these stress the codec, display artifacts readily, and are a significant source of complaints. Second, when choosing a data rate, be sure it’s high enough to deliver artifact-free faces as well as text and other details. Once these components are artifact free, you get dramatically decreasing returns from further bandwidth investment. the PowerMac, though the iMac still lost sync–at all data rates. Since the 800Kbps file produced with the High Profile delivered the same display rate as the 2Mbps file produced using the Baseline Profile, I compared the two files to determine which delivered the best quality. Interestingly, the quality was very similar, thus making the 800Kbps file produced using the High Profile the better option, since it should reduce your total bandwidth costs over the higher-bitrate options. Finally, I tried playing back the file from the local hard drive, as a user would experience if the website offered the file for manual download in addition to progressive download. As you can see, this finally produced a scenario where the file played back smoothly on all three computers. Why would this be? Though the Adobe Flash Player uses the computer’s GPU (graphics processing unit) to play back full-screen video, decoding at 100% of resolution, as I did in these tests, is all performed by the host CPU. In contrast, after downloading the H.264 file, I played it in QuickTime Player, which uses the GPU for full resolution as well as fullscreen playback. In essence, the ability to decode using the graphics hardware makes the QuickTime Player a more efficient playback engine for H.264 video than the Flash player, making download-and-play a good alternative strategy for low-powered computers. When not possible, you’ll get the best results encoded to the lowest possible data rate, which should improve the playback rate on fringe computers. You can view all the files mentioned in this section yourself at the following URLs: • www.doceo.com/H264_Tests/H264_800K/test.html • www.doceo.com/H264_Tests/H264_3M/test.html • www.doceo.com/H264_Tests/H264_2M/test.html • www.doceo.com/H264_Tests/H264B_2M/test.html decoding the truth about hi-def video production Improving the Silverlight Experience To optimize the Silverlight viewing experience, I first tried playing back a file encoded using the VC-1 Main Profile as opposed to the Advanced Profile, which I used in the initial tests. This actually worsened the experience on the PowerMac and had no impact on the iMac. I then tried two alternative data rates, 800Kbps and 3Mbps, finding again that encoding at the lower bitrate improved playback performance on the PowerMac, though not curing issues on the iMac. Clearly, producing at the lowest possible bitrate is the best option for Silverlight. Testing from the desktop revealed the only configuration that played acceptably well on the iMac, again an incentive for HD Silverlight producers to make their files available for download-and-play. The reason is the same as with Flash—that is, neither the Windows nor Mac Silverlight players use the system’s GPU for decoding. Once on the hard drive, the QuickTime Player on the Mac and Windows Media Player on Windows do use the GPU, improving performance, decreasing the load on the CPU, or both. Interestingly, Silverlight’s inability to access the GPU during playback can inadvertently become a problem for producers who use a scaleable player that automatically adjusts the video resolution with the size and position of the browser. For example, the first Silverlight player I 74 STREAMING MEDIA October/November 2008 http://www.streamingmedia.com http://www.cs.ucl.ac.uk/staff/H.Knoche/publications.htm http://www.doceo.com/H264_Tests/H264_800K/test.html http://www.doceo.com/H264_Tests/H264_3M/test.html http://www.doceo.com/H264_Tests/H264_2M/test.html http://www.doceo.com/H264_Tests/H264B_2M/test.html

Table of Contents Feed for the Digital Edition of Streaming Media - October 2008

Streaming Media - October 2008 Table of Contents Standard Time Technology and Business Trends by the Numbers Encoding for Screencams Meeting the Enterprise Video Distribution Challenge Pulling Back the Curtain on Level 3 The 2008 Streaming Media Readers’ Choice Award Winners The Ultimate Guide to Creating Online Video Content That Works, Part 1 Brave New Platforms Video From the Democratic National Convention—Both Official and Unofficial—Made Clear That It’s No Longer Politics as Usual Creative Suite Turns 4 The Future of Internet Radio Decoding the Truth About Hi-Def Video Production Educating the Next Generation of Online Media Makers ViewCast Niagara GoStream SURF The Past and Future of Online Video: I’m Not Dead Yet!

Streaming Media - October 2008

For optimal viewing of this digital publication, please enable JavaScript and then refresh the page. If you would like to try to load the digital publication without using Flash Player detection, please click here.