Prosody Curve

Prosody is the combination of frequency, cadence and intonation that make up the majority of the communicative power of e-language. Chomsky, in his groundbreaking work - defined e-language in conjugate term to I-language , which contacts syntax, grammar and phonemic structure. He simply said it was 'everything else'. Which makes sense, because you want to work with syntax and meaning first, before you get into the frequency distribution. Which is also why most frequency based speech recognizers were early failures.

A good way to think of prosody is to start out inflection. For example, the old rule that a question mark makes the frequency of your voice go up at the end of a sentence. Here's a good paper to read that works for french.

IMHO you should be careful for prosody because sometimes it conveys more information than the original text. I think people tend to discard it. When you prompt an error, make sure that the prompt is self-deprecating if it is the second mistake and automatically fail on the third. In some cases, automatic failure and transfer to live operator should be done on the first. Never take a speech system into active listening for error, it will destroy your chance to build dialogue.

Green Man 2007 this year serves as a focus point to enable me to concetrate on the frequency differences between male and female speech and on the differences between conveyed meaning and emphasis. I will use the exhibit to make sure that where I am going with my research is a good direction. It should be fun. Our entry receipt was confirmed last week, here's hoping we'll get a spot. Hope the dust won't be too bad.

Comments