Yes/No Corpus
                            Release 1.2

              Center for Spoken Language Understanding


UPDATED: 23 August 2002


Directory Structure
-------------------
This document describes the directory structure of this release. 

Following is a written description of the directory structure in 
this release:

  readme.txt	General information regarding the corpus.

  docs/		The documentation directory. This directory
		contains further documentation for the yesno
		corpus.

  labels/	Phonetic labeling directory. This directory
                contains time aligned phoneme-level
                transcriptions (automatic forced alignment).

  misc/		Miscellaneous directory, possibly containing
		software tools and scripts.

  speech/	The speech directory contains the actual 
		.wav files. There are several hundred numbered
		subdirectories within the speech directory.

  trans/	The transcriptions directory. This directory
		contains the orthographic transcription of each
		of the files.

This corpus requires approximately 578MB of disk space.

Visually, the directory structure looks something like this:

				yesno
				  |
	--------------------------------------------------------
	|          |          |          |          |          |
    README.txt   /docs     /labels     /misc     /speech    /trans

The /speech directory contains the speech data.  The files are
divided into sub-directories based on their call number.  Files with
call number 0-99 are in sub-directory "0", files with call number 100-199
are in sub-directory "1", etc.

The /trans directory contains the orthographic transcription of each
of the files.  As with the speech files, the transcription files are
divided into sub-directories based on their call number.  Files with
call number 0-99 are in sub-directory "0", files with call number 100-199
are in sub-directory "1", etc. (A file called trans.txt, containing all
of the transcriptions, is located in the /docs directory.)

Each transcription file looks like:

1,ever_married yes

The first field is a call number call type pair.  This pair uniquely
defines every transcription.  After a single space, the rest of the
characters on the line are the transcription for that file.

Set Description
---------------
The call number for each file is used to determine which set it is in.
We use the call number modulo 5 to determine which set the file
belongs to as indicated by the following table:

	Mod 5 result	set
	-------------------
	0,1,2		training
	3		development test
	4		final test

For example, "YN-1.ever_married.wav" would belong to the training set
since 1%5 = 1.