Yes/No Corpus
                            Release 1.2

              Center for Spoken Language Understanding


UPDATED: 23 August 2002


Overview
--------
The Yes/No corpus is a collection of approximately 20000
examples of people saying "yes" or "no".  This corpus will be
extremely useful for the development of high accuracy yes/no speech
recognizers.

All of the speech in this corpus is telephone speech.

The utterances in this corpus were taken from many other telephone
speech data collections that have been completed at the CSLU.  In many
data collections, the callers were asked to answer various yes and no
questions.  For example, in the Census data collection, we asked each
caller if they had ever been married.  Each of these yes/no utterances
have been gathered together to create this corpus.

Each file in the Yes/No corpus has an orthographic transcription
following the CSLU Labeling Conventions.   In some cases where a
transcription did not already exist we ran the utterance through a
speech recognizer to automatically obtain the transcription.  Since
yes/no recognition isn't perfect, we expect there are some errors in
these files.  However, the error rate is extremly low, so we don't
expect many errors.  We have spot checked the results of the
recognizer and are happy with the results.

Release 1.2 of this corpus contains about 20000 files incuding 
automatic transcription.