Still at the point of being a tech demo, but my latest text-to-screech project has moved forward a little. Here you can see speech sounds controlled by Max 5 piped into PureData. In Pd, I'm using some old tricks with the 'freeze' function in freeverb plus some pitch shifting to further play with the sound. As a potentially interesting wrinkle, the effects in Pd are turned on and off by the words typed in Max: 'reverb', 'freeze' etc.
Hmm. Where to go next?
There's quite a history of musicians and sound artists doing creative things with speech synthesis. One of the best known examples is the Radiohead track Fitter Happier from the album OK Computer, and its not hard to find other cases of commercial artists incorporating this kind of material in tracks.
Very often this has been done on the mac, which has always had speech synthesis built in. (There's a very interesting anecdote about how speech synthesis came to be included on the very first macs at the personal insistence of Steve Jobs.) A number of years ago - I can't find the links now - there was a small community of composers who were authoring and releasing 'tracks' which consisted of nothing but SimpleText files, which were to be 'played back' using the speech synthesis facility. This kind of thing was more effective back then: the earlier versions of the mac speech system responded in interesting and unpredictable ways to abberrant texts.
I've often used this kind of thing in my own work, and I've coined my own term for it: 'text-to-screech'. Here's an example, this is a track called 'vifyavif wif yavif-oo', which also forms part of the instrumental piece donkerstraat:
I've now started work on another such project. This will be a performance piece, where I will be typing text live: I've done work along these lines before, but the new twist will be to try to find away to add extra processing to the speech synthesis live, including perhaps sampling and looping. There are some technical problems with doing this on the mac, however… which I'll make the subject of another post.
I'm working on a new piece towards ARTMUSFAIR/2009, for flute, horn, cello, marimba and 'tape'; this is me working towards the latter, the fixed audio part of the piece. Up to my usual text-to-screech tricks here; this shows some of the tools and methods I use to put things like this together.
Managing to do the same kind of thing now in SuperCollider, speaking one word at a time, which kind of makes more sense in this context; loudcoding?
(Amazed at getting this to work, actually. I'm *so* not a programmer!)
Updated version of the Bare Wires text interface;
Hers is a very early and approximate proof-of-concept video of a possible text-to-screech interface for 'Bare Wires';
The idea is to have a laptop and screen onstage, visible to the audience. The setup is used by various performers during the piece, in various ways; to address the audience, or to direct an improvisation, ask questions, tell a story… anything, really.
'text-to-screech' is my coining for taking familiar text-to-speak technology built into many modern computers, and mangle it, creatively misuse it. The most extensive project I have done along these lines was a commission in 2005 for an online piece for Paragon, which is unfortunately not up any more. A similar strategy was used in 'The Other Other Hand', where creatively edited machine speech was used to represent the voice of the Edwardian composer C. Hubert H. Parry.
The interface shown above is done in Max/MSP, using the built-in voices on a mac. The first aim was to program it so that it would speak each word immediately after it was typed, which was relatively simple to achieve. In addtition, when an 'x' is typed in a word, the partcular voice used changes, typing 'u' or 'v' subtly affects the rate and pitch of the voice. For the next iteration, I want to try the effect of having it speak the word then display it; also to munge the spoken text more drastically, perhaps mutliple voices speaking, perhaps a more clearly pitched approach, perhaps looping a word, so that the result is more 'musical'.
The demo video above fakes up very roughly what it might be like if one performer types up instructions to the others, and then addresses the audience. Another idea I have been playing with is a game whereby the performers are instructed, for instance, to make some sort of distinctive gesture every time an 'a' is typed, and no to obey any other instructions given. So, for instance, the audience sees the performers being told 'play a note'. The performers actually respond to the two 'a's in the sentence by throwing a book to the floor, which is a the prearranged (invisible) instruction. Then the gag is blown by by one of the performers explaining onscreen what is going on. Then another performer changes the rules… etc etc.
(Another, umm, visual/performative reference here is to livecoding, where an artist improvises live onscreen with a computer programming language to produce music and/or visuals, some nice examples here. I've been doing some experimenting in that direction myself using SuperCollider, which can also, as it happens, do text-to-speech. Watch this space…)