From CCNET-L@uga.cc.uga.edu Fri Nov  6 18:39:50 1992
Received: from uga.cc.uga.edu by sirius.ctr.columbia.edu (5.67/5.10) with SMTP id AA23996; Fri, 6 Nov 92 18:39:40 -0500
Received: from UGA.CC.UGA.EDU by uga.cc.uga.edu (IBM VM SMTP V2R2)
   with BSMTP id 3658; Fri, 06 Nov 92 18:35:32 EST
Received: from UGA.BITNET by UGA.CC.UGA.EDU (Mailer R2.08 PTF008) with BSMTP id
 8453; Fri, 06 Nov 92 18:35:30 EST
Date:         Sat, 7 Nov 1992 01:36:02 +0200
Reply-To: "(Tommi V Kaikkonen)" <tommi@PHOENIX.OULU.FI>
Sender: Chinese Computing Network <CCNET-L@uga.cc.uga.edu>
From: "(Tommi V Kaikkonen)" <tommi@PHOENIX.OULU.FI>
Subject:      Idea about spoken language and computers
Comments: To: CCNET-L@uga.cc.uga.edu
To: Multiple recipients of list CCNET-L <CCNET-L@uga.cc.uga.edu>
In-Reply-To:  Ping Zhou's message of Fri, 6 Nov 1992 12:18:23 -0600
Status: RO

Hello there, I am quite new on this list and with chinese language
as well. My idea is to connect computer (Unix or PC) with a speaker
and let it speak any kind of chinese text aloud to me so I can learn
how to pronounce written chinese ... hope so ;-).

I have done one small program which I call gb2pinyin which converts from
Guo-Biao format to tone pinyin. I hope to be able to let my computer
actually talk aloud those tone-pinyin words. My question is: Has anyone
made speech synthesizer capable of doing this for either Sun Sparc stations
or IBM PC compatibles? Any ideas how.. could we for example use some
simple phonetic speech synthesizer and just make one more conversion
filter?

Please respond either to this list or privately to me. Thank you for
your attention.

tommi@phoenix.oulu.fi

From CCNET-L@uga.cc.uga.edu Sun Nov  8 10:31:42 1992
Received: from uga.cc.uga.edu by sirius.ctr.columbia.edu (5.67/5.10) with SMTP id AA05500; Sun, 8 Nov 92 10:31:37 -0500
Received: from UGA.CC.UGA.EDU by uga.cc.uga.edu (IBM VM SMTP V2R2)
   with BSMTP id 8785; Sun, 08 Nov 92 10:28:47 EST
Received: from UGA.BITNET by UGA.CC.UGA.EDU (Mailer R2.08 PTF008) with BSMTP id
 9011; Sun, 08 Nov 92 10:28:45 EST
Date:         Sun, 8 Nov 1992 23:08:22 EST
Reply-To: Carlos McEvilly <UNIT566%TWNMOE10.BITNET@uga.cc.uga.edu>
Sender: Chinese Computing Network <CCNET-L@uga.cc.uga.edu>
From: Carlos McEvilly <UNIT566%TWNMOE10.BITNET@uga.cc.uga.edu>
Subject:      Spoken language and computers
To: Multiple recipients of list CCNET-L <CCNET-L@uga.cc.uga.edu>
Status: RO

This is interesting.  I've been keeping it under my hat, but I've
been working on the same idea and have a prototype working now,
in the testing stages.  Also have a gb2pinyin program, but mine is
big52pinyin (won't use that name, though).

The big5-to-pinyin program should be released in early 1993.

The speech one will take a bit longer.  These run on IBM-PC machines.

- Carlos

From CCNET-L@uga.cc.uga.edu Sun Nov  8 21:57:48 1992
Received: from uga.cc.uga.edu by sirius.ctr.columbia.edu (5.67/5.10) with SMTP id AA10093; Sun, 8 Nov 92 21:57:39 -0500
Received: from UGA.CC.UGA.EDU by uga.cc.uga.edu (IBM VM SMTP V2R2)
   with BSMTP id 0794; Sun, 08 Nov 92 21:54:10 EST
Received: from UGA.BITNET by UGA.CC.UGA.EDU (Mailer R2.08 PTF008) with BSMTP id
 4197; Sun, 08 Nov 92 21:54:08 EST
Date:         Mon, 9 Nov 1992 10:53:57 EST
Reply-To: Carlos McEvilly <UNIT566%TWNMOE10.BITNET@uga.cc.uga.edu>
Sender: Chinese Computing Network <CCNET-L@uga.cc.uga.edu>
From: Carlos McEvilly <UNIT566%TWNMOE10.BITNET@uga.cc.uga.edu>
Subject:      Soundcard
To: Multiple recipients of list CCNET-L <CCNET-L@uga.cc.uga.edu>
Status: RO

Yes, my program requires a soundcard.  I could be done without,
but would not be good enough quality for studying Chinese!

- Carlos

From CCNET-L@uga.cc.uga.edu Mon Nov  9 23:43:06 1992
Received: from uga.cc.uga.edu by sirius.ctr.columbia.edu (5.67/5.10) with SMTP id AA00698; Mon, 9 Nov 92 23:42:54 -0500
Received: from UGA.CC.UGA.EDU by uga.cc.uga.edu (IBM VM SMTP V2R2)
   with BSMTP id 8816; Mon, 09 Nov 92 23:39:27 EST
Received: from UGA.BITNET by UGA.CC.UGA.EDU (Mailer R2.08 PTF008) with BSMTP id
 8853; Mon, 09 Nov 92 23:39:25 EST
Date:         Tue, 10 Nov 1992 12:38:25 +0800
Reply-To: guojin@ISS.NUS.SG
Sender: Chinese Computing Network <CCNET-L@uga.cc.uga.edu>
From: guojin@ISS.NUS.SG
Subject:      Re: Voice Recognition
Comments: To: Chinese Computing Network <CCNET-L@uga.cc.uga.edu>
Comments: cc: Haizi Xu <USERSP45%UBCMTSG.BITNET@uga.cc.uga.edu>
To: Multiple recipients of list CCNET-L <CCNET-L@uga.cc.uga.edu>
In-Reply-To:  Your message of "Mon,
              09 Nov 92 11:11:26 PST." <9211091914.AA26712@iss.nus.sg>
Status: R

>     Synthesized Chinese speech is very interesting to me. But I
> am more interested in voice recognition as an input method for
> Chinese word processing. I'd like to know about it. Any info is
> welcome.



To my knowledge, among others, the following three products
on market in China are the most well-known:

(1)  Star-852 from Beijing's STAR Co.
(2)  Xinghe-1 (?) from Foshan's Xinghe Co.
(3)  THED-919 from Beijing's CNED Co.

Star-852 and Xinghe-1 were first introduced in late 1989 or early
1990 while THED-919 was in late 1991.  They are similar in
following aspects:

o  golden/silver medals winners
	at national and/or international exhibitions, some have
	won several times.
o  integration
	provide not only speech recognition, but also speech
	synthesis (generation), optical character recognition,
	sentence level intelligent input methods supported by
	natural language understanding techniques, enhanced word
	processing and powerful desktop publishing (formatting,
	printing, etc.) as options.
o  expensive
	in the range of 10,000 to 20,000 RMB (approximately
	US$2,000 - 4,000)
o  low compatibility
	due to, among other reasons, the needs of special
	hardware (recognition card, synthesis card,
	microphone, speaker, publishing support card, ...),
	Chineselization (han-hua), and non-standarization,
	you are recommanded to check if it fits your environment.
o  isolated speech
	which means you need to read to the computer
	hanzi by hanzi (not word by word!)
	with clear pause/stop in between.
o  speaker dependent
	which means before you can use the recongition system,
	you need to teach/train it, that is, to read a
	predefined material under certain conditions.
	It usually takes hours.
o  90-95% accuracy
	although it varies person to person, if cooperative
	and patient enough, you can achieve even higher.
	(specially trained operators can achieve virtually
	100 percent accuracy with a few 'tricks' at public
	demonstrations).  On the other hand, if you have strong
	accent and/or pronounce carelessly, the accuracy can
	easily drop to lower than 50%.
o  20-90 hanzi per minute
	it's much slower than that a specially trained
	operator usually achieves in coding-input and
	only a little bit faster than that of pinyin-input.
o  high-tech
	the three are all integrations and commecializations
	of different research prototypes which were constructed
	at the Beijing's topmost Tsinghua University and
	intensively supported by several China's national
	high-tech projects.

It is said that Star Co. has agents in states but I have only
the THED's ads in hand.  You may contact them at:
	Mr. Wang ZhengXian (Director)
	Department of speech recognition
	China National Electronic Devices Corp. (CNEDC)
	#A-65, FuXingLu Road,
	Beijing, 100036, P.R.China
	Tel:  86-1-8211166
	Telex: 222932 CNEDC CN
	Fax:  86-1-8217517	(they prefer FAX comm.)

Also, there are many pre-market prototypes in-demo in dozens of
institutions and universities which are currently conducting
Chinese speech recognition research.  Two among the most
successful are:

(1)  Tsinghua University, Beijing
(2)  National Taiwan University, Taibei


Yours
GuoJin

