Names Corpus
                            Release 1.3

              Center for Spoken Language Understanding


UPDATED: 23 August 2002


Overview
--------
A common problem in training and developing speech
recognition systems is scarcity of data, especially
particular phonemic contexts.  The Center for Spoken
Language Understanding is attempting to address this
problem with the Names Corpus.  The Names Corpus is a
collection of name utterances, both first and last names,
from several thousand different speakers over the
telephone.  Name utterances are "spontaneous" in that the
subject is not reading from a word list.

Another area of active research is the development of name
Recognition systems.  The Names corpus is a useful resource
for addressing this problem.

The utterances in this corpus were taken from many other
Telephone speech data collections that have been completed
at the CSLU.  In most data collections, the callers were
asked to leave their name at some point.  Also, the callers
would occasionally leave their name in the midst of another
utterance.  The names in these situations were extracted
out of the host utterance and added to the Names Corpus.

Each file in the Names corpus has an orthographic
Transcription following the CSLU Labeling Conventions.
Also, to take advantage of the phonemic variability, many
of the utterance have been phonetically transcribed.  The
selection of files to phonetically transcribe was
constrained by a process that selected files that were
suspected to contain phonetic contexts that had not yet 
been transcribed.

Release 1.3 of this corpus contains 24245 files.  All of 
these, have been phonetically labeled.  Approximately 40% of
the bigram phonemic contexts possible, without regard to
language constraints, are represented.

Description
----------
There is a large variability in the spelling of English
names.  In the case of common names, plausible spellings
were intuitive.  However, for the rarer names, we
transcribed using an orthography which resembled the
pronunciation as closely as possible.  We have not
attempted to standardize the name spellings.

Over the whole corpus there are about 10570 unique names.
No standard spellings are used so names such as "kerri" and
"kerry" will be counted as two separate tokens.

The corpus consists of about 6.3 hours of speech.

The following table gives a count of the number of files 
for each utterance type.

	Type		Number
	----------------------
	firstname        9727
	lastname        11431
	other1		  151
	other2		   29
	other3		    2
	other4		    1
	whole		 2659

Recording Conditions
--------------------
Each subject called the CSLU data collection system by
dialing a toll-free number. Depending on which data
collection the caller was calling, the call was recorded
over an analog line, or a digital line.

The analog telephone line was connected to a Gradient
Technologies box. Data from incoming calls were recorded by
the Gradient box. The sampling rate was 8khz and the files
were stored in 16bit linear format on a UNIX file system.
Each utterance was recorded as a separate file.

The digital data were collected with the CSLU T1 digital
data collection system. The sampling rate was 8khz and the
files were stored in 8-bit mu-law format on a UNIX file
system.

Subject Population
------------------
Subjects whose utterances are included in this corpus are
respondents to USEnet postings, radio advertisement,
newspaper advertisements, and interoffice memos.

Annotation
----------
All of the files included in this corpus have corresponding
non-time-aligned word-level transcriptions that comply with
the conventions in the CSLU Labeling Guide. In addition,
conventions in the Labeling Guide.