Video quality matters

by S J Birkill, Archive Editor

What's wrong with this picture?

YouTube has revolutionised our access to what we used to call television, but in so doing it has lowered the baseline of what most consider acceptable picture quality. Every step on the road to high-fidelity viewing was hard won. The first experimental broadcast images were built up from 30 vertical lines stacked horizontally to produce a portrait-shaped (3:7 ratio) moving image. 240-line broadcasts followed, but the BBC's first 'high definition' service, begun in 1936 from Alexandra Palace with 405 horizontal scan lines (two interlaced fields, each of 188.5 active lines) was the first to offer comfortable family viewing in the living room, and continued to do so (with an interruption for WWII) until the mid-1980s. Incremental improvements had progressed in parallel since the '60s: 625 lines, then colour, then wide screen: the old squarish 5:4 and 4:3 aspect ratios gave way to a cinema-style 16:9 format. The 1990s brought us digital TV, at first by satellite and cable and then over the air, and digital delivery meant that true HDTV could become the norm: 625 lines (now 'standard definition': 576 active lines in its digital form) gave way to 720, 1080, or as many as 2160 picture lines — 'HD' and '4K'. Waiting in the wings is '8K'.

Digital TV resulted in a convergence of TV technologies with those of computers and the Internet, and YouTube, established in 2005, became the default site for television content delivery via the Web. Initially conceived as a platform for individuals to share their home video 'footage', YouTube's success soon led to online video becoming commercialised, with new advertising, donation and subscription-supported channels appearing daily. And the company's relaxed attitude to copyright (they block content only after receiving complaints from the rights-holders) made it a repository for historical creative art from the television and film world that would otherwise have been lost to the public, perhaps for decades awaiting legal resolution of rights issues; perhaps for ever.

And this is where clivejames.com comes in. When Clive started his website in 2004, delivering TV content over the Web was an expensive business: video servers were costly to run, and full-resolution 576p digital TV programmes consumed large chunks of storage and bandwidth, all of which had to be paid for at the prevailing per-megabyte and per megabit-per-second rates. The on-demand service needed to cope with an unpredictably large number of simultaneous streams to prevent the viewer's experience stuttering to a pixellated halt, and this multiplied the bandwidth requirement and thus the cost. Clive's Internet ambition was set commendably high, but the project's budget wasn't quite up to it.

There was also the user's end of the link to consider. Today, with fibre to the home ubiquitous and 5G mobile coming up fast, we've become blasé about downstream bandwidth: 100Mb/s is nothing special any more. But in 2004 most homes had an Internet speed measured in low single figures of megabits per second. Full-resolution video wasn't viable for them, demanding more than the available bandwidth. YouTube was on the horizon, ready to handle the 'transmission' end, but the 'last mile' bottleneck was still there. The solution was to discard information — to use digital compression technologies at source in order to reduce the occupied bandwidth.

One way to achieve this is by halving the horizontal and vertical resolutions: a PAL (625-line) picture of 768 x 576 becomes 384 x 288 pixels, a reduction of 75%. But the default settings on video software at the time would assume source material in the American NTSC (525-line) format, with a picture size of only 640 x 480 pixels. So the 'half-resolution' version would measure 320 x 240 pixels. You can see where we're going — the UK ditched 240-line TV broadcasting in 1937! Yet this resolution is still widespread on YouTube, its apologists arguing that it's 'only as bad as' VHS or Betamax. But in fact it's much worse! A Betamax recording will lose some of its horizontal definition and about 50% of the original's colour resolution, and the video noise spectrum will change, but all 625 lines are still there, in 1:1 correspondence with the original — the VCR's analogue processing makes a convincing show of 625-line material. In contrast, the digital down-sampling inherent in the 320 x 240 process needs to employ heavy anti-alias filtering in the spatial domain, both horizontally and vertically, to remove sampling artefacts created in the down-scaling process: its output is a small and very fuzzy rendition of the original. I could continue, and discuss frame-rate reduction, also employed to cut bandwidth, before we even begin to employ a lower sampling frequency or to apply the proprietary compression techniques inherent in the now obsolescent Flash and RealMedia coding formats. But I won't — the image above might suffice.

Apart from video compression, it's easy to screw up the aspect ratio when you don't know exactly what format the original used, or whether it should be measured in square or rectangular pixels. Sky TV went through a period of using the intermediate 14:9 frame size for its analogue wide-screen transmissions, as a compromise between the standard and wide-screen frames. Needless to say, most viewers saw either a stretched, cropped or squished version of the picture. Those who habitually stretched everything to fit their wide-screen screens probably didn't notice, or didn't care. But if you're uploading to YouTube it's essential to get it right, or your viewers will see the distortion, or black bars at top and bottom of the picture ('letterboxing') or side bars ('pillarboxing'). Or all three, as shown above. [For this archive I have chosen, where possible, to embed the videos within our standard layout — this gives us some control over how the material is presented, and avoids much of the distracting advertising and unsolicited promotions YouTube is prone to. The YouTube controls are still available in the window, so 'Full Screen' display can be invoked if desired.]

It's not just resolution and aspect ratio: gamma, brightness, contrast, hue and saturation are also there to be messed with. And what about sound? The mutilations seen and heard on YouTube continue to defy belief. Some might be explained by multiple generations of digital tweaking: if the original version is lost and only an early downsampled YouTube copy remains, it's tempting (now that bandwidth is less of an issue) to apply further processing to up-scale the image or to correct distortions. The outcome is seldom good. Lip-sync is easily caught in the crossfire: in the absence of noise-reduction buggerage, downsampling the audio (say from 320kHz at 48 or 44.1 kbps to 32kHz at perhaps 11.025 kbps) is usually safe, though with muffled results, but audio timing is frequently a casualty of sequential video processing operations.

As clivejames.com launched its Video section before YouTube really got going, decisions had to be made about video file size and the number of streams the site could handle. Talking in the Library was first priority, and the first few shows to go online, produced by Welcome Stranger on a Chinese takeaway budget, were streamed in either Flash or RealMedia format from a dedicated video server, until it became apparent to Clive that the process could not sustain further expansion. As luck would have it, on-line magazine Slate stepped in and offered professional hosting services, as Clive details in earlier versions of his Video Gateway page.

In 2019, when I undertook to build this archive, I discovered the extent of disarray surrounding the residual video copies of the series. All that I had to work with was what remained on YouTube. I couldn't identify the point where Slate's hosting had ceased, and what had happened to Clive's copies of the original material. Certainly YouTube carried versions of most of the 40 programmes I'm aware of, but the quality varied widely — some I considered unwatchable, and those weren't always the ones from the first three seasons, shot in 4:3 aspect ratio. Every one of them showed at least one of the video encoding mistakes I've mentioned here — short, fat, fuzzy orange Clives were everywhere. Most of the episodes were under YouTube accounts bearing Clive's name. One of these accounts was opened in December 2017, so this may have been the point at which Slate pulled out, leaving Clive's webmaster of the time to maintain continuity by uploading them to YT, apparently without access to the originals. A second 'Clive James' account appeared in early 2019, during the downtime of clivejames.com, and duplicated some of the content from the earlier account, plus new episodes from the later seasons. Some of the shows were in a less mangled condition the second time around. It seems neither of the uploaders had pristine copies of any of the programmes, nor access to each others' work. A few episodes appeared on YT under other users' accounts, and these may have benefitted from being recorded off air, because they looked much better — they avoided the look of having been twice through the wrong mill, backwards.

The accompanying audio was not safe from mutilation. From the start, two microphones were used, the interviews being recorded as dual mono to allow balance adjustment and any muting during post-production. I don't think any were recorded in true stereo, though two of the programmes I've included in the AUDIO section here reveal that Sky mixed them to 'ping-pong' stereo, separating the voices by panning them left and right. This yielded an unnatural mock-stereo effect, so I've used the mono mix for all those audio versions. Many of the lowest-resolution videos also have audible coding artefacts on the soundtrack, and appear to reveal the presence of RealMedia digital compression, a process both audio and video may have been subject to at one time. The Flash format, standard in the early days of YouTube, had also been used. The Season 5 shows, with two exceptions, have been passed down to us with their sound grossly mangled, with a 'flanging' effect superimposed on the digital compression artefacts, and the occasional momentary drop-out unrelated to Sky's muting of the odd profanity. The exceptions — versions of the Stephen Fry and Victoria Wood interviews uploaded by independent users — may have been captured off-air and so avoided these distortion processes.

The final series, Season 6, was recorded in 2008 in conjunction with TimesOnline, apparently for release exclusively on the Web. This has fared worst of all: I found no complete episode had survived on the Internet. The Times produced trailers using very short clips of the Tom Stoppard, Germaine Greer, Will Self and Patrick Stewart interviews, which appeared in the second of the 'Clive James' YouTube channels with very poor image quality; of the Salman Rushdie and Melvyn Bragg programmes I have yet to find a trace. Only one of the earlier episodes remains AWOL: the Alexei Sayle chat from Season 5, which I seem to recall having watched some ten years ago but regrettably didn't record or download at the time.

My aim here, naturally enough, is to present the material at its best. Clive has made excuses for picture quality, blaming the meagre production budget of the early seasons, but the inadequacies in that department are now barely perceptible, submerged beneath the layers of well-intentioned but misguided processing and re-sampling applied between the master tapes or drives, and the Web. With a lot of work I could correct the lip-sync, volume, EQ, hue, saturation, brightness/contrast/gamma and aspect ratio errors in all but the most extreme cases (as I have done for the accompanying screen shots), but I can't reconstitute the lost information. For now, I could cling to the hope that the redoubtable Claerwen James will discover a hoard of Betacam tapes or DVD-R's among Clive's belongings (I suspect his Web-artist-of-the-day had access to decent versions before the steamrollers moved in), or that Sky will raid their archives and cough up their own full-resolution originals. Or perhaps some fan home-taped or DVD'd all the broadcasts on first transmission...