Skip to content

Imaginary Machines: eMC

October 14, 2010

All right, so you get a bunch of party people, and they take turns being Selector and MC.

The Selector controls a poetry generator, which is producing lines for as long as a music loop is going. The poetry generator has ePoGeeS-like control of phoneme and number of syllable preferences as well as texts being used, all of which can be updated during run-time to change the poetry being currently produced. The poetry generator could have some kind of interface through which the Selector chooses among many possible candidates. The Selector also decide which music loop is currently being played; a music loop could be as simple as a cut-up like I did below, or range to consumer-grade riddims, or hip-hop or techno beats or the like. The music loop should be simple and repetitive; not as complex as karaoke versions of songs, for example.

The MC controls a Text-to-speech engine: speed, tone, possibly intonation, but most importantly: mute. The MC can decide to recite the lines (either in whole or in part) instead of allowing the Text-to-speech engine to say them; the MC could decide to ignore the generator’s output entirely.

Meanwhile, the Party People are giving feedback on how they like what’s going on, saying “yeah,” “right on,” “uh-huh”, etc. A Speech Recognizer and Classifier interpret whether their input is generally positive or negative. (The Selector keeps an eye on the Speech Recognizer and Classifier, and trains it to the feedback style of the Party People.) The feedback is used to influence the poetry generator: lines that received positive feedback are more likely to be selected for during a stochastic beam or evolutionary search. The Party People are also responsible for deciding when the Selector and MC are through and who should replace them.

I think it’d only take the spare time of a couple days to hook something like this up (except for the speech recognition and classifier, but that could be replaced with a directional mike and volume tracking if need be.) But for some reason this system doesn’t really interest me. On the other hand next time I do some TTS or ASR I might take a couple minutes to do something for this. Dunno, maybe a 2 in 10 chance it’ll get finished any time soon, maybe a 6 in 10 chance it’ll get done eventually. Method 9059622c-1514-40b7-b89c-ee208bba9634 (eMC).

Advertisements
No comments yet

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

%d bloggers like this: