background speech generator

Idea emerging from discussions with my girlfriend:

Suppose you are sitting in the bus, overhearing some people speaking in a foreign language. You don't understand a word, but you still listen to the melody of the language, you observe some patterns, the distribution of certain sounds, you make guesses about the topic from the intonation or the emotions expressed and your imagination runs freely.

Then you start talking with someone in your own native language that most of the others in the bus don't understand but you notice them starting pay attention, probably in the same way as you did earlier. You are wondering, how does it sound, what is the melody of the language, what images do the sounds evoke, but you can't really tell, because as you listen to your own native language, you actually understand it and the meaning overrides any free association.

The solution to this "problem" would be a kind of "Muzak" (see link) for natural language. A computer program that given enough samples as training data can generate random speech that has all the same properties as the original speech, probably same syllables with the same frequency, smoothly blended in random order, similar intonation of sentences, same melody of the whole speech, but without making any sense whatsoever. That way you could listen to the melody of your native language without paying attention to the meaning.