means that the data rate can be reduced by a useful amount. Codecs that offer this feature include CELP and the Qualcomm PureVoice. The compression ratio alters the slope of the input/output transfer characteristic. The compressor will also have a gain control, as a compressed signal will sound louder than the uncompressed. The limiter prevents the output level exceeding a preset threshold. Attack and release times The dynamic range controls do not change gain instantaneously; that would just modulate the sound with its own level. Instead there is a delay before the control parameter is changed the attack time and another delay as the applied parameter is restored the release time. The attack time is customarily very short, just a few milliseconds. It is long enough for brief transients; the limiter takes care of them. The release time is much longer, up to several 204 The Technology of Video and Audio Streaming 0 -20 -40 -60 -60 -40 -20 Output level (dB) Input level (dB) unity gain amplification attenuation 4:1 compression peak limiter Figure 10.9 Compressor limiter transfer characteristics.
Note: If you are looking for good and high quality web space to host and run your application check Lunarwebhost PHP Web Hosting services
levels will change. During pauses in the presentation you do not want to amplify the background noise. Peak limiting Simple clipping generates unpleasant audible artifacts. For the best results you should use a look-ahead peak limiter. They reduce the gain fractionally before a peak, gracefully reducing peak levels without the apparent distortion that you get with straight waveform clipping. The amount of peak limiting you use depends upon the program content. If you want to retain musical transients to give punch to the audio the limiter should be used with caution. Noise reduction A noise gate reduces gain radically below a preset threshold. One application that can be very useful is to stop the dynamic range compressor from sucking up noise during pauses in a presentation. It should not be used if the background is contributing to the audio track; an example would be background music. If the noise gate were to cut the music in and out this would be irritating to the listener. The only way to set the threshold for the noise gate is to listen to the effect. The threshold can be increased gradually until the bac ground noise cuts out. If there is unacceptable breathing of the background, it can be backed off until an acceptable result is achieved. The noise gate usually has temporal adjustments, so that there is a delay before it cuts in and out. If the gate is maladjusted, it can lead to clipping of the presenter s speech. A noise gate offers a great advantage if your compression codec has silence detection. During periods of silence the compressor effectively shuts down. This Preprocessing 203 20 50 100 200 500 0 -20 1k 2k 5k 10k Frequency [Hz] A]Bd[ l evel oi du presence boost bass roll-off Figure 10.8 Voice equalization.
Note: If you are looking for good and high quality web space to host and run your application check Lunarwebhost Cheap Web Hosting services
is the luxury of more time than you are likely to have for a webcast. Always apply equalization with care; too much will not enhance the final result. For streaming, typical adjustments could be as follows: Low-pass filter to remove high-frequency hiss High-pass filter to remove hum and air conditioning rumble Mid-frequency mild boost to give presence to speech You may find problems with sibilance from some speakers. It can sometimes be reduced with a microphone shield; otherwise, a combination of EQ and multiband compression can assist. Dynamic range Modern home audio systems can reproduce a very large dynamic range. DVDA and SACD both offer 24-bit coding; home theater systems also offer a wide dynamic range. In contrast, streaming audio often is listened to with the limited facilities built into laptop PCs. A loudspeaker, maybe 1 inch in diameter, driven with a modest power amplifier, cannot hope to reproduce a wide dynamic range. So, much like AM radio, intelligibility is improved if the dynamic range is compressed. Two processes are employed: compression and peak limiting. The compression ensures that the average loudness of the signal is relatively constant. This requires more sophisticated processing than a simple automatic gain control. The more complex dynamic processors split the audio into bands, typically five. This band splitting was developed for the original AM radio processors, to give the most natural sound with high levels of dynamic range compression. By using phase-linear filters for the band splitters, the final mix will be free of nonlinear group delay. If you are going to host corporate webcasts, usually the executives will address an audience from a podium. As the speaker moves about, the audio 202 The Technology of Video and Audio Streaming 28 0 dB - 24 dB 24 dB 56 113 225 450 900 1.8k 3.6k 7.2k 15k Figure 10.7 Graphic equalizer.
Note: If you are looking for good and high quality web space to host and run your application check Lunarwebhost Cheap Web Hosting services
Equalization Often called EQ, equalization is seen everywhere from car audio to the recording studio. Equalization allows you to adjust the frequency response of the audio chain. The simplest equalizers are bass and treble controls. Equalizers can be used to compensate for deficiencies in the original recording, and to allow for limited audio replay facilities of the average PC. Bandpass filters These give comprehensive control over frequency response with a number of bandpass filters, often called graphic EQ from the graphic layout of the controls. Parametric equalizers The most flexible filters allow the bandwidth and center frequency of each filter to be adjusted. They are used more often in the recording studio, where there Preprocessing 201 picture cadence cut same different different 1 2 3 1 2 1 3 2 1 frame ‘n’ frame ‘n+1′ frame ‘n+2′ frame ‘n+3′ frame ‘n+4′ 30 frame/second interlace scan picture cadence fields in a frame same different different same same same 1 2 3 1 2 1 2 3 1 2 1 3 2 3 2 3 same different different 1 2 3 1 2 1 3 2 cadence disrupted edit point Figure 10.6 Editing the telecine cadence.
Note: If you are looking for good and high quality web space to host and run your application check Lunarwebhost PHP Web Hosting services
giving two interlaced fields. In the case of film this would give a field rate of 48, rather than the 60 of video. To correct the field rate, alternate frames are scanned three times; that is, the first and third fields are duplicated. This gives a 3:2:3:2 sequence to the scanning so that 24 frames become 60 fields, the normal rate for video. Note that a video frame can be made from fields from two different film frames. If this video signal is to be encoded efficiently, the duplication of one field every five is redundant information that could be removed. This task is not as simple as it appears the repeated field carries no label to indicate the duplication. The duplicate field has to be detected in the presence of analog noise (from the telecine image sensor). This noise makes the duplicated fields slightly different. The second problem is that if video transfer of the film is edited, the 3:2:3:2 sequence can be disrupted. So that means that once a duplicate field is detected, the next duplicate cannot be assumed simply by counting fields. This process of removing the duplicate fields is called inverse telecine. 625/50 systems completely avoid this process. The film simply is run slightly fast, at 25 frames per second, and scanned directly to 50 fields in a normal interlace sequence. The 4 percent increase in speed is ignored. Audio Most audio content will need some preprocessing before the streaming codec. This will include equalization, dynamic range compression, peak limiting, and noise reduction. 200 The Technology of Video and Audio Streaming discarded field frame ‘n’ frame ‘n+1′ frame ‘n+2′ frame ‘n+3′ frame ‘n+4′ discarded field 24 frame/second progressive scan 24 frame/second film 30 frame/second interlace scan 1 231 2 1 2312 Figure 10.5 The telecine sequence.
Note: If you are looking for good and high quality web space to host and run your application check Lunarwebhost PHP Web Hosting services
Inverse telecine and 3:2 pull-down When film is transferred to video (a process called telecine), to maintain the correct television timebase, the film shot at 24 frames per second has to be processed to 30 frames per second. Normally each frame is scanned twice, Preprocessing 199 white black Median filter Brick wall filter Original Figure 10.4 Median versus brick-wall filter with impulse noise.
Note: If you are looking for good and high quality web space to host and run your application check Lunarwebhost Adult Web Hosting services
the advantages of the median filter is that it does not soften the edges of an object. Brick-wall filter A brick-wall filter allows both the horizontal and vertical spatial resolution to be filtered extremely sharply above a preset frequency point. This lets you reduce high-frequency detail out of band detail that cannot be resolved in scaled-down pictures at low bit rates. Typically film grain and electrical noise have a high frequency spectrum, so this filter will remove both. To avoid the softening of edges, the brick-wall filters can apply a boost just inside the pass-band. Note that a brick-wall filter will smear out impulse noise, so the median filter should be applied first. Recursive filter Recursive filters are used to remove the random noise from film grain or electrical noise. For typical video material the main change from one frame to the next is noise. A recursive filter averages one frame with the next, reducing the noise but leaving repeated picture information unchanged. The only picture information that does not repeat should be moving objects within the scene. A motion detector can be used to separate real moving objects from noise. The recursive filtering can be made adaptive, and can be applied solely to the stationary areas of the picture. This avoids the smearing of moving objects if too much filtering is applied in an attempt to clean up a noisy picture. Sharpening Sometimes a picture needs sharpening. This is difficult without adding noise to the edges of objects. Once a picture has become soft, maybe through excessive dubbing, information has been lost forever. If you are careful a picture can be given a bit more snap, but beware of using too much sharpening. Unsharp masking Many video processors offer a technique called unsharp masking, or USM. This is a traditional technique to sharpen edges in an image that has been borrowed from film compositing. A thin, soft negative is sandwiched with the original positive. When a new exposure is made it sharpens the edges without increasing noise in flat areas. It produces a more pleasing effect than a straight sharpen. 198 The Technology of Video and Audio Streaming
Note: If you are looking for good and high quality web space to host and run your application check Lunarwebhost Adult Web Hosting services
Noise reduction Noise comes in many different forms. It can be random, caused by electrical noise or film grain. Film-originated stock also has scratches and dirt. Tape is not without its problems videotape suffers from tape dropouts. FM noise causes sparkles on satellite contribution links. If your material has already been compressed and then decompressed, there may be visible block errors. Although they are artifacts rather than noise, cross-color and cross-luminance from composite video sources also eat into your bandwidth budget. All these forms of noise require different treatment, although some noise is difficult to remove satisfactorily without introducing a different set of artifacts. For example, very heavy filtering can produce motion artifacts; this will cause smearing of fast-moving objects. There are three popular types of filters for cleaning up video: the two spatial filters, median and the brick wall; and the temporal recursive filter. Median filter These are used for the suppression of impulse noise, such as tape dropouts or switching glitches. Impulse noise is usually black or white. A threshold detector separates the impulse noise from the picture information. A median filter then replaces the noise pixels with the median (not the mean) of a window of surrounding pixel values. It is also called a sliding window spatial filter. One of Preprocessing 197 CRT mask scan raster safe action area safe text area Figure 10.3 Safe text area.
Note: If you are looking for good and high quality web space to host and run your application check Lunarwebhost Adult Web Hosting services
The original tube television receivers had poor voltage regulation for the EHT supply to the CRT. This meant that the picture size was inversely proportional to the picture brightness. To mask this change in size, the picture was overscanned; that is, the raster was larger than the CRT display area. This also covered up poor scan geometry and drifts with age. So the viewer saw only the center 90 percent of the picture. Aware of this, the cameraman would insure no important action took place at the edges of the picture. Any graphics also are designed to sit well within the actual display area. This is called the safe text area. Media players display the entire picture, so the edge usually lost on a television is also displayed. This means that a picture can be cropped to the safe action area before encoding with no loss of useful picture information. If material is shot specifically for streaming, then the camera operator can crop in closer. 196 The Technology of Video and Audio Streaming lift black white gamma black white gain black white Figure 10.2 Lift, gain, and gamma.
Note: If you are looking for good and high quality web space to host and run your application check Lunarwebhost Adult Web Hosting services
fields, one with the even numbered lines and one with the odd (in NTSC, fields are captured every 16 ms). To generate the progressive frame the two fields are combined. Unfortunately if there are any moving objects in the scene, the result will be a horizontal streaking or feathering effect. A de-interlace filter smoothes the two fields to give an average position for the object. Most codecs offer a de-interlace filter to improve rendition of fastmoving objects. The filter is required only for interlaced video. Cropping One easy way to cut down the amount of data is to crop the picture. Surely you want to encode the entire frame? Well, the edges of the frame rarely contain useful picture information. Preprocessing 195 16 0 235 255 black zero white peak level sample value television computer video 0 255 zero (black) peak level (white) sample value sync pulse Figure 10.1 Video levels.
Note: If you are looking for good and high quality web space to host and run your application check Lunarwebhost Adult Web Hosting services