DARPA Resource Management Continuous Speech Database 
                                    (RM1)
 
                     Speaker-Independent Training Data 
                               NIST Disc 2-3.1
 

This speech database is a collection of recordings (corpus) of spoken
sentences pertaining to a naval resource management task.  The corpus consists
of two primary components:  one component, which was designed for
speaker-independent speech recognition research, system development and
evaluation, consists of relatively few utterances from each of 160 subjects,
while the second component, designed for speaker-dependent speech recognition
studies, consists of many utterances from each of 12 speakers.  The corpus has
been further divided into portions for training, development testing, and
final evaluative testing. 
 
Subjects read the sentences from written prompts in low background noise. 
The material was recorded using a Sennheiser HMD 414 headset microphone and
simultaneously digitized at a 20 kHz sampling frequency with 16-bit
quantization.  The digitized speech data was downsampled to 16 kHz and
segmented into files corresponding to individual sentence-utterances.  Most
of these sentence-utterances are approximately 3 to 5 seconds in duration. 

All of the sentences are consistent with a limited language model that allows
queries about ships, ports, etc., along with commands to control a graphics
display system, but little else.  There is no "official" language model, but
a crude (non-probabilistic) word-pair grammar that provides complete coverage
of the sentences in this corpus is included.  
  
This CD-ROM (NIST disc 2-3.1) comprises all of the speaker-independent
training data for this corpus.  This disc contains 3360 NIST-headered speech
sphere files as well as several documentation files.  Sentence text prompts
are included but "official" transcriptions (orthographic, phonetic, etc.) do
not exist and have, therefore, not been included.  For the purpose of system
testing, it has been assumed that the prompts represent an accurate
orthographic transcription of the utterances.

This disc is the second in a series of CD-ROM-based speech and language
corpora being prepared by the National Institute of Standards and Technology
(NIST) and distributed through the National Technical Information Service
(NTIS).  Development and preparation of this database was made possible by
support from the Defense Advanced Research Projects Agency (DARPA) Information
Science and Technology Office.  

Speaker-Independent Training Material
-----------------------------------
This disc contains the following speech data for each of 80 speakers: 

       2 dialect calibration sentences (files sa[1-2].sph)
      40 speaker-independent training sentences (files from sr[001-600].sph
                                                      and st[0001-2235].sph
      --
      42 total sentence utterances per speaker


CD-ROM Directory Structure
--------------------------
The CD-ROMs' directory hierarchy is structured so that a full path/filename
uniquely identifies an utterance (the database, data usage, speaker, and
sentence).  Note: Identical filenames DO EXIST and, therefore, care must be
taken when copying files from the CD-ROM to other media.  However, if the
original path/filename is lost, a file may be disambiguated by the utterance
identifier in the header.

CAUTION:  Eight of the 80 speakers designated for speaker-independent training
material have also been used as test speakers in previous DARPA/NIST benchmark
tests.  These speakers' files have been placed in a separate directory,
"excluded", and should NOT be used in speaker-independent system training. The
remaining 72 speakers are the "official" DARPA speaker-independent training
speakers.

The directories are structured as follows:  

<DATABASE-FILE> ::= <SPEECH-FILE-SPEC> | <DOC-FILE-SPEC> | <README-SPEC>

  <SPEECH-FILE-SPEC> ::= /<DATABASE>/<USAGE>/<SPEAKER_DIR>/<SPEECHFILE>

    <USAGE> ::= ind_trn | excluded

    <SPEAKER_DIR> ::= <SPEAKER_ID>_<DIALECT>
      <SPEAKER_ID> ::= adg0 | ... | wem0
      <DIALECT> ::= 1 | 2 | ... | 8

    <SPEECHFILE> ::= sa1.sph | ... | st2235.sph

  <DOC-FILE-SPEC> :: = /<DATABASE>/doc/<DOCFILE>
    <DOCFILE> ::= documentation files and directories  (see below)

  <README-SPEC> ::= /<DATABASE>/readme.txt

<DATABASE> ::= RM1


Example directories and files:

/                              (CD-ROM root directory)
rm1/                          (database identification)
rm1/ind_trn/                  (speaker-independent training data usage)
rm1/ind_trn/adg0_4            (speaker "adg0", dialect region "4")
     .
     .
rm1/ind_trn/wem0_5            (speaker "wem0", dialect region "5")
rm1/dep_trn/adg0_4/sa1.sph    (speech sphere file containing an utterance
     .                          of sentence "sa1")
     .
rm1/ind_trn/adg0_4/st2165.sph (speech sphere file containing an utterance
                                of sentence "st2165"
rm1/excluded/                 (excluded speakers who have also been used in
                                benchmark tests)
rm1/doc/                      (online documentation and tables - see below
                                for file descriptions)
rm1/readme.txt                (this file)


Online Documentation
--------------------


The following documentation files can be found in the rm1/doc directory:

     al_sents.snr - Complete listing of all RM Database sentences in SNOR
                    form.  SNOR, an acronym for Standard Normalized
                    Orthographic Representation, is a uniform way of writing
                    the words and sentences in this corpus.  SNOR-format 
                    sentence texts are required as reference material for the
                    DARPA/NIST standard scoring software.
     al_sents.txt - Complete listing of all RM Database sentences in prompt
                    form.
     it_scrpt.txt - Table listing the speakers on this disc and the scripts
                    of sentences they read.  The scripts are located in the
                    the rm1/doc/scripts directory.
     it_sents.snr - Listing of sentences on this disc in SNOR form. 
     it_sents.txt - Listing of sentences on this disc in prompt form.
     it_spkrs.txt - Table describing the speakers on this disc.
     header.def   - NIST header object definitions.
     lexicon.snr  - Complete Resource Management lexicon in SNOR form.
     not_used.txt - List of valid RM sentences not used in the corpus. 
                    (These sentence texts were generated, but never recorded.)
     scripts/     - Directory containing lists of sentence identifiers.
                    Each list (script) indexes the order of sentences
                    spoken in one recording session by a particular speaker.
     wp_gram.txt  - Resource Management Word-Pair Grammar (developed at BBN).


NIST Header Structure
---------------------
This series of CD-ROMs employs a new NIST speech file header structure. The
header is an object-oriented, 1024-byte fixed-length, entirely ASCII
structure.  The header is composed of a fixed portion followed by an
object-oriented variable portion.  The fixed portion is as follows:

NIST_1A<new-line>
   1024<new-line>

The first line specifies the header type and the second line specifies the
header length.  Each of these lines are 8 bytes long (including new-line) and
are structured to identify the header as well as allow those who do not wish
to read the subsequent header information to programmatically skip over it. 

The remaining object-oriented variable portion is composed of
object-type-value "triple" lines which have the following format:

<LINE> ::= <TRIPLE><new-line> |
           <COMMENT><new-line> | 
           <TRIPLE><COMMENT><new-line> | 

  <TRIPLE> ::= <OBJECT><space><TYPE><space><VALUE><OPT-SPACES>

    <OBJECT> ::= <PRIMARY-SUBOBJECT> | 
                 <PRIMARY-SUBOBJECT><SECONDARY-SUBOBJECT>

    <PRIMARY-SUBOBJECT> ::= <ALPHA> | <ALPHA><ALPHA-NUM-STRING>
    <SECONDARY-SUBOBJECT> ::= _<ALPHA-NUM-STRING> | 
                              _<ALPHA-NUM-STRING><SECONDARY-SUBOBJECT>

    <TYPE> ::= -<INTEGER-FLAG> | -<REAL-FLAG> | -<STRING-FLAG>

      <INTEGER-FLAG> ::= i
      <REAL-FLAG> ::= r
      <STRING-FLAG> ::= s<DIGIT-STRING>
      
    <VALUE> ::= <INTEGER> | <REAL> | <STRING>  (depending on object type)

      <INTEGER> ::= <SIGN><DIGIT-STRING>
      <REAL> ::= <SIGN><DIGIT-STRING>.<DIGIT-STRING> 

    <OPT-SPACES> ::= <SPACES> | NULL

  <COMMENT> ::= ;<STRING>  (excluding embedded new-lines)

<ALPHA-NUM-STRING> ::= <ALPHA-NUM> | <ALPHA-NUM><ALPHA-NUM-STRING>
<ALPHA-NUM> ::= <DIGIT> | <ALPHA>
<ALPHA> ::= a | ... | z | A | ... | Z
<DIGIT-STRING> ::= <DIGIT> | <DIGIT><DIGIT-STRING>
<DIGIT> ::= 0 | ... | 9
<SIGN> ::= + | - | NULL
<SPACES> ::= <space> | <SPACES><space>
<STRING> ::=  <CHARACTER> | <CHARACTER><STRING>
<CHARACTER> ::= char(0) | char(1) | ... | char(255)

The currently defined objects (used in this database) are listed in the file
rm1/doc/header.def.  (Note: The list of objects in header.def may be expanded
for future databases, since no order or number of objects is imposed on this
header structure.  The file header.def is simply a repository for "legal"
object definitions.)

The single object "end_head" marks the end of the active header and the
remaining unused header space is undefined.

The following is an example header from the Resource Management database:

NIST_1A 
   1024 
database_id -s3 RM1 
database_version -s3 1.0 
utterance_id -s8 adg0_sa1
channel_count -i 1 
sample_count -i 50074
sample_rate -i 16000 
sample_min -i -2032
sample_max -i 2708
sample_n_bytes -i 2 
sample_byte_format -s2 01 
sample_sig_bits -i 16 
end_head 

A document will be forthcoming which describes the header structure in greater
detail and basic "C" software modules will be available in the future
for header interpretation, generation, and manipulation.