LINGUIST List 5.851

Thu 28 Jul 1994

FYI: Speech database, AAAS, Software for modeling sound change

Editor for this issue: <>

Directory

1. Malcolm D. Crawford, new speech database to be recorded in September 1994
2. Vicki Fromkin, aaas
3. , Software for Modeling Sound Change

In September, researchers from Sheffield University, England, will join
colleagues here from ATR to record a new speech database.
There are a number of novel aspects to this work, and this message is being sent
at this time so that anybody who wishes to do so can express their interest
 and/or
offer suggestions or advice.

Background
Research in the Hearing Group at Sheffield University focuses mainly on Auditory
Scene Analysis (ASA), or "the separation of sound sources" -- e.g. extracting
speech from a mixture of speech and noise. In order to test our models we
require a database consisting of multiple simultaneous sources (several acoustic
sources are present at the same time), in "real world" conditions (i.e. not an
anechoic chamber!): this clearly contrasts with most speech databases collected
to date!

The corpus we collect may, however, be of interest to researchers in other
fields e.g.

 * sound localisation
 * speaker identification
 * source segregation
 * dialogue/discourse analysis
 * intention processing
 * prosodic analysis
 * spoken language understanding
 * automatic speech recognition
 * virtual reality

as will become apparent in the following brief descriptions of the recording
environment and proposed scenarios.

Recording environment

The database will be recorded in the acoustically controlled room at ATR, with
reverberation etc. commensurate with an average-sized meetings room.

The room impulse response will be measured.

There will be up to five (5) participants in any single session.

Up to eight (8) channels of data will be recorded: 1 for each of the speakers,
1 from a centrally-placed omnidirectional microphone, and 2 binaural from a
B&K mannikin.

Recordings will be made using an 8-track digital tape recorder (TASCAM DA-88)
at 48kHz (for final distribution tracks will be downsampled to 16kHz).

A video recording of the session will be made.

A small amount of speech data will be collected from each speaker individually
(for use in ASR research etc.).

Scenario

Five people will engage in a task-oriented discussion. The task will be chosen
to provide not only the likelihood of much overlapped acoustic material, but
also to be a realistic topic of discourse for researchers in Natural Language
Processing.

Recordings will be made of a number of tasks: one will be collaboration on
solving a crossword. We welcome in particular suggestions for task domains of
interest to researchers in other fields (a colleague from the United States has
suggested recording a rehearsal of an a capella singing group, which would be
of interest to the Music community!)

Analysis

For our work we only require "light" annotation, indicating: which speaker was
talking at any time (a word-level analysis is sufficient); any intrusive noise
sources, and areas of overlap between sources. Given the excellent facilities
at ATR we will almost certainly be able to also provide a phonemic
transcription. Others have expressed interest in producing prosodic analyses,
and we would welcome offers of any other form of analysis.


The above description is deliberately brief. Please feel free to contact me,
either formally or informally, personally or by Email, if you have any interest
in this work. We would be pleased to receive advice from the experts here at
ATR, and to try to accommodate other's wishes for this database.

I look forward to hearing from you,

Malcolm Crawford.

Currently at:
 mmalchip.atr.co.jp
ATR Human Information Processing Research Labs.
2-2 Hikaridai, Seika-cho Soraku-gun, Kyoto 619-02, Japan
Phone: +81-7749-5-1089 Facsimile: +81-7749-5-1008

A reminder -- the AAAS (American Association for the Advancement of Science)
which publishes Science weekly and which is the major umbrella organization
for all scientific disciplines in the U.S. and which holds an Annual
meeting now has accepted linguistics as a science by establishing Section Z
-- on Linguistics and the Language Sciences. In order to get final approval
in the next couple of years we must get 400 to 500 members of the AAAS to
designate Section Z as their affiliated section. We are far short of that.

So -- I am writing to urge all linguistic scientists (and that means
everyone on the Linguist Net) to join AAAS and so designate Section Z
(if you haven't already).You can FAX your membership application to
(202) 842 1065 if you wish, asking them to charge your Visa or MasterCArdgiving
 the number and expiration date) or asking them to Bill you later if you
are joining as a Regular member. The membership fees are:
 for U.S. Residents: Regular -- $92. Postdoc $67 Student $50.

or you can mail your membership fee or send your credit card number to:
 AAAS, PO Box 2033, Marion OH 43306-2133

The next meeting will be held in Atlanta Georgia, from Feb 13-19. Section
Z is sponsoring 3 symposia plus jointly sponsoring with Psych and Speech
two others. When the program is out I will send another message to Linguist.

We look forward to your membership and your attendance at the next meeting.

Vicki Fromkin, Secretary, Section Z

 My IBM-compatible program, PHONO, for operating models of regular
historical sound change, has just been made available for downloading by
anonymous FTP from the Online Linguistics Archive at the University of
Michigan.
 The user writes sound-change rules in PHONO's notation and supplies
ancestor words. PHONO applies the rules and shows the derivation.
 The program is accompanied by a model for Spanish.
I am eager to find out if the program works equally well with models
for other languages.
 Here's how to get the program by FTP:
 -- Type the following:
 ftp linguistics.archive.umich.edu
 -- When asked your identification, type the following:
 anonymous
 -- When asked your password, type your e-mail address.
 -- When you have gained access to the Archive, type the following:
 cd linguistics/software/dos
 -- Before downloading, set the mode to binary, by typing the following:
 binary
 -- Finally, type the following:
 get phono.zip
 You will receive the files in "zipped" (i.e. compressed) form.
Use PKUNZIP software (available from your local installation) to
decompress them.
 For more information, contact me.
-----------------------------------
Lee Hartman ga5123siucvmb.bitnet
Department of Foreign Languages ga5123siucvmb.siu.edu
Southern Illinois University
Carbondale, IL 62901-4521 U.S.A.