Search company Blinkx is happy that it has landed a deal with ITN to automate the process of making video clips. It's happy because it's a nice high profile deal to have in a nice high profile part of the market – mobile and IP TV – but also because this is the first time the company's sold anything. Until now, it's been trying to get people interested by giving away a desktop search tool and talking about contextual analysis of the Web, but to be honest there are lots of people doing both those things and my first experience of the Blinkx desktop and Internet search tools was unimpressive enough to be my last.
That stuff's still going on, but barring some whacky deal I doubt it's going anywhere. I said as much to Graeme 'Scoop' Wearden as he packed himself off to a meeting with the company, but the video stuff turned out to be fun enough.
On his return to the office, labouring under what might fairly be called mild enthusiasm, I went to have a look. I tried a few search terms, and it worked surprisingly well. But we all know what new search engines are best tested with – one's own name (not for nothing is Google an anagram of Ego Log) and naughtiness. I elected to try marijuana.
The first few hits were the usual small crime reports and off-beat "Hey, they smoke pot in Amsterdam!" reports. But then some cat called Mourinho appeared, who is apparently in charge of some association soccer outfit. I watched the clip: there was no mention of the evil weed. I asked Graeme; he phoned Blinkx — and now we know.
It's very difficult to search video, because computers are hopeless at recognising what they're looking at — and by hopeless, I mean at the level of telling the finest algorithms in the world that "There is a chair, an apple and a man with a beard in this picture, and nothing else. Where are they?" in order to get a result. So, you have to cheat, either by having thousands of bored students watching video clips and typing in what they see — which doesn't work — or by listening to the sound track.
Which is what Blinkx does. It breaks down the sound track into a string of phonemes, which it then matches to words with similar sequences of sounds. Hence amusing Mourinho/Marijuana mixup. And when you think about it, it is perfectly possible to make a large number of rude words from non-rude sounds — the hilarious Norfolk Enchants county slogan, for example.
But they'd better be careful: if a backing musician who went on a folk tour with a famous female singer was careless enough to say "Yes, I was KD's fiddler", they stand a chance of being pulled up by those looking for kiddyfiddler stories — and that sort of thing sometimes occasions a sense of humour failure and slander by association writs.
It's a fun site, though. Give it a go — and do pass on any particularly fun phoneme phumbles.