I've been experimenting the last few days with feeding back speech recognition interpretations of a paragraph from HG Well's The War of the Worlds back into speech recognition systems to see what effect they had. I'm using a system called 'talkback', which is part of the Microsoft Speech Development kit. I'd open several talkback windows, read a few sentences and the let the system speak back to itself over and over.
I'm a little unsure what to make of my experiments. I came to using speech recognition systems creatively with a mind to exploring what James Guetti calls 'sentence sounds' - based on the premise that, as speech recognition systems look at the words surrounding a word to calculate what it is, they might go for a sentence in a way that might be at least usefully dis-analogous to how we might grasp a sentence aesthetically.
But what my play has thrown up is a decaying discourse, where the only constants: "I heard" before it gives its interpretation and "when you said "after" get increasingly repeated. It gets stuck in one distortion of these - for instance "when you said" gets transformed into "when you shouldn't" "when you used" etc. Occasionally it throws up something suddenly strange and poignant (for instance, "and we want to understand" in one long piece of ramble.
Clearly my methodology is not appropriate to my original interests, but I wonder if I've stumbled upon something of potential interest here? I'm not sure if there's any direct knowledge or insight to be had; but perhaps it could be manipulated to serve as an interestingly dis-analogous model of some other process? Something of Bernard Stiegler pops into my head - was it 'circuits of trans-individuation' perhaps? I'll have to take a look.
In any case, I hope to mix a track out of it for the next Fuselit, so hopefully I can make something enjoyable to listen to at least.
No comments:
Post a Comment