What is required to produce the best video
with Microsoft Windows Media?
All live streaming encoders utilize Microsoft's streaming video algorithms.
But Microsoft does not allow access to the inner workings of those algorithms.
Consequently, to improve the quality of streaming media, one must "optimize"
the media before passing it to the software encoder engine.
Compression engines provide amazingly reduced data rates from those required by normal broadcast media. Uncompressed broadcast-quality video requires 20.995MB/sec. [Mega-Bytes per second] or 167.961Mbs [Mega-bits per second]. Microsoft's software compression algorithms provide compressions as great as 5831:1 at 28Kbs [28,000 bits per second] to as little as 82:1 at 2.0Mbs [2 megabits per second]. These dramatic reductions are accomplished via several processes, including image scaling, motion estimation, temporal and spacial compression.
Live encoding requires that all processes be completed in realtime. If the time required to accomplish any process can be reduced or eliminated, that time is then available for the remaining processes. Consequently, if one allows all the available processing time to be spent on final compression rather than image scaling, the resulting images will exhibit fewer compression anomalies or artifacts.
Equipment Choices
Since the quality of the original video plays the major role in determining
the quality of the resultant streaming signal, it is of primary importance
to utilize the best quality equipment practical for your application.
When selecting video and encoding equipment, remember that the resultant
images will only be as good as the weakest link in the process.
Cameras
Today's technology allows excellent images to be reproduced by relatively
inexpensive cameras. The image quality previously found only in the $30,000+
video cameras of a few years ago can today be approximated with cameras
costing under $5,000. And, high-quality 3-chip cameras can now be purchased
for under $2,000. Slight increases in image quality are apparent in more
expensive cameras up to approximately $25,000. Select the best 3-chip
camera practical for your application.
If your program format allows, the best quality stream will be created by routing your cameras output directly to a high-quality encoder. If your format requires editing and the addition of other media before final stream creation, an intermediate recording device must be used to first capture and edit the program segments.
Intermediate Recording Devices
If you use a non-linear editing (NLE) system to perform your editing,
you can achieve the least reduction in picture quality by recording directly
into that system from your camera without first recording to videotape.
Choose an adequate digitizing rate on your NLE to capture all the quality
from your camera. Some NLEs allow video to be digitized at uncompressed
data rates (21MBs), which assures that there is no loss of original picture
information. However, such data rates require massive, high-speed hard
drives to store the captured signals. Most users find, given a high-quality
input signal directly from a camera, digitizing rates of 6-10MBs are adequate
to capture all of the video information without introducing appreciable
noise into the signal.
If your program format requires that you first record your video to a camcorder or VCR before editing, use the best quality VCR practical for your application. Use the following popular VCRs in the listed order of preference to capture the best quality video:
1) 4:2:2 Digital VCRs; Sony DigiBeta, JVC
Digital-S, & Panasonic DVC-Pro50 ($7,000-$35,000)
2) 4:2:2 Analogue VCRs; Sony BetaSP ($9,000+)
3) 4:1:1 Digital VCRs; Sony DV-Cam, Panasonic DVC-Pro25, & any MiniDV
($1,500-$7,000)
4) S-VHS VCRs; from Panasonic, Sony, JVC, & others ($400-$3,000)
5) Hi-8 / VHS / 3/4" VCRs; from Panasonic, Sony, JVC, & others
($80 up)
6) 8mm VCRs; from Sony, Canon, & others
When capturing video into your NLE from any of the above listed VCRs, be sure to select an adequate digitizing rate to reproduce the original quality. As with cameras, most users find that, given a high-quality input signal from a VCR, digitizing rates of 6-10MBs are adequate to capture all of the video information without introducing appreciable noise into the signal. However, if your source information is noisy (seen as a grainy or pasty image), you may have to increase your capture rate to 10+MB/s to prevent the introduction of additional noise into that signal. Video noise does not compress well under any compression algorithm, including MJPEG and MPEG.
Internet Encoders
As with cameras and VCRs, select the best quality encoder practical for
your application. Encoders that are specifically designed for high-quality
streaming will provide a superior Internet image as opposed to those devices
that are designed for other applications. Today's professional encoders
are available from EarthCaster,
Digital Rapids
and others and range from approximately $5,000-$30,000. Professional units
incorporate broadcast-quality analog-to-digital & digital-to-digital
converters, broadcast-quality frame buffers, video noise reduction, and
hardware scaling. They also provide the capability of simultaneously creating
multiple streaming datarates in realtime and are bundled with a customized
versions of the Windows Media encoding engine.
If these units are not practical for your applications, any Windows-compatible video capture card can be used with the downloadable version of Microsoft Windows Media Encoder to create one live stream or streaming file. If you are using a Firewire (IEEE-1394) camera or VCR and your encoding PC also has a Firewire input, you may not need a video capture card at all. As long as your CPU has a clock speed of least 2.0Ghz, your system should be able to encode directly from your Firewire source.
If you are using a video capture card, certain output signal formats from your camera, VCR or NLE will provide higher quality video with a better signal-to-noise (S/N) ratio than others. Always select the highest quality output format from your source and match it with the highest quality input format that your encoder offers. For best encoding quality, select the following I/O format in the listed order of preference:
1) SDI digital / Component analog
2) IEEE-1394 / Firewire digital
3) Y-C / S-VHS analog
4) Composite analog
If you already have your video and encoding equipment, there are ways to optimize their quality.
The three most important optimization techniques
are...
1) Increasing Signal-to-Noise Ratios
2) Hardware Pre-scaling
3) Signal Pre-Processing
Signal-to-Noise Ratios
Signal-to-noise ratio is a relationship between the portion of data that
is actual audio or video information [signal] and the portion of data
that is spurious information [noise]. A higher signal-to-noise (S/N) ratio
is experienced as better quality audio and video.
S/N ratios are measured in decibels (dBs) for both audio and video. Each 3dB increase in the S/N ratio decreases the noise in a signal by 50%! With that in mind, you can see that a relatively small increase in S/N can dramatically increase audio or video quality. All media compression algorithms work much better when dealing with high S/N ratios. Again, noise does not compress well and creates artifacts in a streaming signal.
You can optimize the S/N of a camera or
VCR by following these guidelines:
1) Virtually all of today's better
video cameras have better S/N ratios than virtually any VCR. Recording
any camera signal to videotape adds noise to the original signal. If possible,
it is preferable to use the signal directly from a camera (or audio mixing
board) as the input source for your encoder, rather than first recording
that signal to digital or analog tape and then encoding that signal.
If recording your program information first is necessary (in order edit and add additional media and graphics), use the best quality recording device available (one with the highest S/N ratio). Whether you record to a VCR or directly to a non-linear editor, use your camera's best output signal format as the input format for your recording device. Use output formats in the following order of preference.
1) SDI digital / Component analog
2) IEEE-1394 / Firewire digital
3) Y-C / S-VHS analog
4) Composite analog
2) Adequate lighting is also important to optimize the S/N of any camera. While some cameras will produce better images under low lighting conditions, all cameras produce better images under ideal lighting conditions. If possible, make sure that your subject matter is well and evenly lit, without hot spots or dark areas. Selecting the luminance "boost" or "gain" of +6dB, +12dB or higher on your camera will also add noise to your video. If possible, do not adjust the gain on your camera.
3) If your camera's or VCR's output is less than ideal due to its inherent quality, or poor lighting conditions, you may route the camera's or VCR's output signal through a "video noise reduction" device before sending the signal to an encoder. These devices are designed to filter out some of the video noise in your signal. Certain units can reduce noise levels up to 75%. Video noise reduction circuitry can be found in many of today's high quality frame synchronizers and TBCs as well as in stand-alone noise reduction devices. ($2,500-$15,000)
4) Tape can also be an important factor in recording quality. Always use the best quality videotape. The small amount of money saved by using a lesser-quality videotape can cost you a considerable loss in S/N.
Hardware Scaling
Full resolution CCIR-601 video is 720 horizontal pixels by 486 viewable
vertical lines (720x486) in a 4:3 aspect ratio. Most web video is only
320x240 pixels or smaller. The scaling of 720x486-pixel video to 320x240-pixel
video immediately eliminates more than 78% or the original image information.
The challenge of an almost 80% size reduction while attempting to maintain
image quality is further compounded by the fact that video pixels are
a different aspect ratio than computer pixels.
Broadcast video pixels are not square, but are tall, narrow rectangles. Computer pixels, however, are square. Consequently, scaling a 720x486-4:3 image to 320x240-4:3 image requires the pixels to be converted from rectangular to square. In addition, the color of each resultant pixel must be calculated from the surrounding pixels in the original image to assure the most accurate representation.
This scaling may be done via software, but to accomplish this process in the best way requires considerable calculation... typically more processing than can be accomplished well in realtime with most of today's CPUs. High quality scaling can, however, be accomplished quite efficiently via hardware computer chips specifically designed for that purpose. Pre-scaling video to the desired stream size via hardware before it is passed to the software encoders will result in higher stream quality.
If your encoder allows for hardware scaling, be sure to activate it. If your encoder does not offer this option, make sure that you employ the highest speed computer practical. Dual CPU units are also better suited for encoding than single-CPU units.
Signal Pre-Processing
Video
In traditional broadcasting, the FCC ordains video signal voltages and
parameters, determining what is "broadcast legal" and what is
not. These parameters have no bearing on streaming media, however. One
may adjust your incoming video precisely to broadcast standards and yet
receive less than ideal streaming video.
All encoding algorithms have some less-than-desirable effects on both video and audio. Of course, when compressing a signal to less than 2% of its original data rate, this is to be expected.
With most encoding algorithms, the black level of an original video signal seems to be lowered 3-5% causing the streaming video to appear darker than the original. The Gamma of an image can also be affected, causing the stream to appear somewhat "washed-out", lacking appropriate contrast.
If your encoder offers a "Processing Amplifier" (ProcAmp) feature to allow adjustments to video parameters, raise the black level and/or video level slightly and select or adjust the Gamma settings to optimize the resultant stream. Some encoders offer adjustments both for the incoming video and outgoing stream. You may also use an outboard ProcAmp or one that is built into a frame synchronizer or TBC to adjust the signal before it reaches the encoder.
Make small adjustments and then create a 30-second streaming file. View that file with the appropriate player. If the file still does not look right, make additional adjustments and repeat the process. If you are monitoring the video before it gets to the encoder, you may now notice that your incoming video looks too bright or has too much contrast. But remember, only the quality of the resultant stream is important.
Audio
Audio compression also affects the signal in order to attain the dramatic
data rate reduction necessary for streaming. Very high and low frequencies
are suppressed by the compression algorithms. This can result in audio
that sounds somewhat muffled and lacks dynamic range.
There are a wide range of audio processing devices that can preprocess the signal to help compensate for these anomalies. Parametric Equalizers, NoiseGates, Compressor/Limiters, Normalizers, DeEssers, BassBoosters etc. can be used to adjust audio parameters to allow the resultant stream to sound more like the original. Again, if you are monitoring the audio before it gets to the encoder, it may sound strange. Remember to monitor the output stream with an appropriate player when making audio adjustments. Use the same small-step-by-small-step process that you used to adjust the video signal while optimizing the audio signal.
By compensating for the effects caused by software encoding algorithms, one may produce streaming media with better quality video and audio. By adjusting the black level, video level, saturation level, hue and gamma values, one is able to "normalize" streaming video. By adjusting the EQ, adding noise gates, compressor / limiters, normalizers and DeEssers on the original audio signal, remarkably better streaming audio can be produced.
Unless your content is focused on audio, you will experience higher quality if you select Mono rather than Stereo. Stereo dilutes the sampling rate by 50% at a given datarate.
Data Rate, Framerate and Pixel Size Selections
The goal is steaming is to obtain the optimal video and audio image at
a given data rate. Otherwise, one wastes bandwidth... and bandwidth equates
to money. If your original video was shot with a standard 4:3 aspect ration,
make sure that your stream size is also 4:3; ie. 640x480, 480x360, 400x300,
320x240, 160x120 etc. If your original video was shot in a 16:9 aspect
ratio, make your stream 16:9 also; ie. 640x360, 480x270, 320x180, etc.
The encoding algorithms also work better and more efficiently if you select
a pixel size and data rate that is equally divisible by 16. For example,
a 320x240 stream with 16Kbs audio and 128Kbs video will look and sound
noticably better than a 321x241 stream with 18Kbs audio and 130Kbs video.
For high-action, fast changing video scenes, it will be important to select 30 frames per second (fps). However, if your presentation is mostly talking heads, use 15fps. The video will be appreciably clearer at a given datarate.
Custom Profiles
Creating a custom, optimized Windows Media profile for your streams will
provide the best audio and video quality possible at the lowest bandwidth.
Though this process, users can get the most for their streaming dollars.
The "sweet spots" for encoding
most content will use the following custom encoder settings:
Talking-Head Type Video with embedded
streams for both dial-up and broadband
Size: 320x240 pixels
Audio Bit Rate: 16-bit / 16kHz mono
Audio Codec: Windows Media Audio 9
Video Codec: Windows Media 9 Video
Video Bit Rate (1): 28Kbs
Video Framerate (1) : 10 fps
Quality Slider (1) (Smoother/Sharper): 60
Video Bit Rate (2): 128Kbs
Video Framerate (2) : 15 fps
Quality Slider (2) (Smoother/Sharper): 70
Talking-Head Type Video with embedded
streams for broadband
Size: 320x240 pixels
Audio Bit Rate: 16-bit / 16kHz mono
Audio Codec: Windows Media Audio 9
Video Codec: Windows Media 9 Video
Video Bit Rate: 128Kbs
Video Framerate: 15 fps
Quality Slider (Smoother/Sharper): 70
Action-Oriented Video for broadband
Size: 320x240 pixels
Audio Bit Rate: 20-bit / 32kHz mono CBR
Audio Codec: Windows Media Audio 9
Video Codec: Windows Media 9 Video
Video Bit Rate (1): 240Kbs
Video Framerate (1) : 30 fps
Quality Slider (1) (Smoother/Sharper): 70