The Switchboard corpus consists of telephone conversations between
unacquainted adults. The corpus contains 2430 conversations averaging
six minutes each, and totaling 240 hours of speech, or 2.4 million
words.  The particpants were 543 speakers (302 male, 241 female) from
all areas of the United States.

A computer-driven "robot operator" system handled the calls, giving
the caller appropriate recorded prompts, selecting and dialing another
person (the callee) to take part in a conversation, introducing one of
about 70 topics for discussion and recording the speech from the two
subjects into separate channels until the conversation was finished.

The train directory contains a careful transcription of the speech
data by humans.  One would like to train a speech recognition system
to come as close to this transcription as possible.  

The test directory contains a bunch of candidate transcriptions
produced from the speech by an existing speech recognizer.  The final
step of speech recognition is to choose among these candidates using
an n-gram model -- that's your job!