100 utterances is the standard LANGUAGE minimum but there are few recent references for how many syllables, words or utterances for analysis of connected SPEECH
It depends on your purpose. If you are doing research on a particular type of sound, then you have a great deal of freedom in how long the passage is.
On the other hand, if you want a passage that represents all of the relevant sounds or sound combinations, then you'll probably need something at least as long as the "North Wind and the Rain" passage. Rupal Patel and Kate Connaghan have been working on such a passage for children, "The Caterpillar", which seems to be near to being published:
http://www.cadlab.neu.edu/publications.php
I'm sure they could answer your question more specifically.
In the aphasia literature, I see ranges from "6 minutes" to "12 minutes" as well as "200 hundred utterances" The trick is to balance adequate sample length and keeping it manageable for transcribing and coding. Spontaneous speech is tricky too....the emotional valance of the topic can vary the sample a person provides.