Encoding Best Practices, Part 1

 Continue Reading

Encoding Best Practices, Part 2

This month I take a break from our HD production chores to look at streaming-related issues. In this first installment, I''ll share the results of a survey I recently completed that details the best practices of high-profile broadcast and corporate streaming producers, examining options such as codec choice, video resolution, audio and video data rates, and frame rate. In the next issue, we''ll compare the quality of the big three codecs—H.264, VC-1, and VP6—using SD, HD, and screencam videos.

About a year ago, I was speaking at a seminar on codec usage and decided to mix a bit of practice in with my theory. That is, I could talk my heart out about codec quality or the optimal encoding parameters for streaming, but actual market usage seemed much more compelling than my own unsupported theory.

So I grabbed a couple of programs that shall remain nameless and started capturing and analyzing streams of video from well known broadcast and corporate sites. I''ve updated the statistics twice since then, and I will present my most recent statistics, captured in May 2008, in this article. For the record, note that I only reported on files that I could actually download and analyze, which excluded multiple sites.

Nonetheless, the broadcast sites that I reviewed included ABC News, Accuweather, ABC, BusinessWeek, The Weather Channel, The New York Times, Forbes, Sports Illustrated, ESPN, CNET, ETonline, The Golf Channel, DL.TV, CBS (show replay), The Wall Street Journal, CNN, Fortune, MSNBC, and CBS (show preview). On the corporate side, I looked at PricewaterhouseCoopers, infoDev, PI Worldwide, Deloitte, EMC, Accenture, E&Y, Chevron, HP (case study), Cisco, Intel, HP (corporate TV), EDS, Microsoft, and and GE.

I chose the broadcast sites for obvious reasons. On the corporate side, I chose high-profile organizations that made videos fairly easy to find on their external websites. Note that many organizations included multiple codecs on their sites. For broadcast sites, I analyzed the video files most prominent on their home page. For corporate sites, I analyzed the most recent videos that I could find on their sites.

Overall, I don''t claim that these results are statistically meaningful, but they do illustrate general trends and best practices.

Table 1. Codec usage by broadcast and corporate sites tested.
Click here for a larger image

Let''s start with codec usage which is shown in Table 1.

Clearly, in the broadcast space, Flash had taken over, growing from 50-percent to 88-percent usage. In the corporate space, however, Windows Media remains strong, actually growing in usage since the November 2007 survey.

Some interesting factoids that showed up during my testing. First, the only time that I saw QuickTime and Real were when they were offered as an alternative to Windows Media Video files (by Accenture and PI Worldwide). Neither were ever the sole alternative. Another interesting point was that multiple companies, including Microsoft, used both Windows Media and Flash, although never as alternatives to each other.

Finally, none of the Flash producers used the H.264 codec, although two broadcast producers were still using the Sorenson Spark codec. The rest, of course, were using On2 VP6.

Table 2. Output parameters by category.
Click here for a larger image

Next, I wanted to examine the resolution and data rate of the video streamed by the world''s largest organizations, which is critical for several fairly obvious reasons. First, if CNN is streaming video at 576x324 at a combined data rate of close to 600kbps, this tells me that most viewers can retrieve and smoothly stream at least this data rate. Second, it also tells me, assuming my target demographics are similar to CNN, that my target viewers expect streaming video to look as good as that of CNN.

Table 2 shows these results, which I''ve segmented into three categories for broadcast and two for corporate with the numbers in parentheses in the Category column showing the number in each category. I include total pixels in the chart because that''s the best way to compare the video sizes among the categories. For example, at 201,984, total pixels in the Broadcast – Aggressive category are almost three times those of the Broadcast – Conservative category, while those of the Corporate – Conservative category, at 71,314 pixels, are less than half those of the Corporate – Aggressive category.

Table 3. Increase in data rate over 13 months.
Click here for a larger image

Table 3 shows how total data rates (video and audio) have increased over the last 13 months. Clearly, a substantial group of producers assume that their viewers can comfortably receive more than 600kbps, a dramatic increase since April 2007. Clearly, if you haven''t re-evaluated the output parameters of the files you''re streaming in the last year or so, now would be a good time.

Table 4. Data rate at given frame rates.
Click here for a larger image

Next, I wanted to determine how the frame rate used by the various entities impacted the data rate. In theory, one reason to use 15fps as opposed to 30fps would be the ability to drop the data rate. Certainly this is true in the corporate space, where those producing at 15fps used a data rate 89 percent lower than those producing at 30fps.

In the Broadcast – Midrange group, which was the only broadcast group with a meaningful sampling of 15ftp and 30fps producers, the difference wasn''t quite that stark. Those producing at 15fps used an average data rate of 25 percent less than those producing at 30fps.

Interestingly, however, those producing at 24fps used a lower data rate than 15fps producers. If you have the option to shoot at 24fps, this may be a technique that allows you to drop the data rate while retaining both smoothness and visual frame quality. Note, however, that all producers in the Broadcast – Aggressive group produced at 30fps.

Table 5. Audio data rates for Flash and Windows Media Audio.
Click here for a larger image

The final parameter I reviewed was audio data rate, looking at two specific items. First, many producers use 128kbps audio just because it''s a default in many templates. Clearly, if the Broadcast – Midrange group thinks that 66kbps is fine for Flash and 48kbps fine for Windows Media Audio, it''s probably sufficient for most producers.

I separated these stats out by Flash and Windows Media to highlight the fact that the MP3 codec used in Flash is less efficient than either Windows Media Audio or the AAC audio codec included in H.264. In fact, some broadcast producers still using Windows Media, such as Sports Illustrated, broadcast at 32kbps. If you''re planning on switching from Windows Media to Flash, you'd better count on bumping the same data rate to produce the same quality.

That's it for this issue. Check back next time for a comparison of H.264, VC-1, and VP6 codecs using SD, HD, and screencam videos.