Hi Garardo,
I am not by any means versed in the science of sound, but I understand
that
words are seldom if ever actually pronounced exactly the same way, even if
they ostensibly appear to be the same. The effect, I believe, is called
co-articulation. A particular sound is influenced by the surrounding
words/phones.
Keith
-----Original Message-----
From: owner-spk2wrt@phoenix.edc.org [mailto:owner-spk2wrt@phoenix.edc.org]
On Behalf Of spk2wrt@phoenix.edc.org
Sent: Saturday, February 05, 2005 4:48 AM
To: spk2wrt@phoenix.edc.org
Subject: Re: Speech recognition system?
[MOderator's note. Thanks, Gerardo. I think there are LOTS of issues in
using speech recognition. I've done a lot of work with English speakers
who have non-standard speech. I assume the issues for people trying to use
other language versions of DNS with speech differences in those languages
are similar, if not exactly the same. I know from personal experience that
it was nearly impossible to enroll someone with heavily-accented English
in the English version of DNS a couple of versions ago. I don't know if
this is Asim's area or not, but it certainly is important. Bob]
Asim,
I was waiting eagerly for a message like yours. I
recently read some writings from Richard Stallman
(Free Sofware Foundation) about free software. While
being a supporter of the fundamental ideas of the FSF,
I had never understood how was that Linus Torvalds a
supporter of the GNU-copyleft license in his Linux
software did not endorse completely the politics of
Stallman. Somehow this book allowed me to understand
that Stallman was stubborn in his idea of getting an
"already made" piece of software to assemble as the
kernel of the FSF unix clone. Some other ideas of
Stallman was his encouraging to reverse-engineer
software in order to speed the development of free
software.
As far as I have seen, in this list what is basically
discussed is the application of the DNS speech
recognition system. Some time ago sombody wrote saying
that she was trying to use the DNS with a blind person
with a dysartria. My friend, employer, partner here in
Madrid has cerebral palsy with a severe dysartria.
Anyway, we made a hard try with DNS.
I would like to make some considerations. As far as I
understood, the DNS has two modules: one of speech
recognition (this combination of sounds make this
word), and one of language recognition (this
combination of words make this sense); each one with
its specific "made in a English speaking nation"
weight. I mean, in Spanish (with a very few exception)
every letter from the alphabet can sound always the
same. One of the problem of the cerebral palsy
dysartric persons is that they can not reproduce
similar sounds easily.
My complaints with the DNS system would be of the very
very few parameters that the user can modify. In the
case of my friend (we worked with less than ten words
--individually and making short phrases with them)it
was frustrating tha the linguistic-prediction module
introduced a lot of "noise" predicting the same fame
few words again and again. From the point of view of
Stallman, a system is not free if you can't see its
source code. In this case, I know that DNS is a
copyrighted system, but it was frustrating for me to
acknowledge tha a piece of sofwtare like MS-Word has
scores of parameters available to common people
(secretaries, you and I) and DNS has very few. In our
failure-try I would have liked to be able to diminish
the prediction module, in order to make more weighty
the speech (not language, meaning) recognition module.
Besides, it is frustrating to Spanish speaking users
the idea that Spanish is almost 100% "phonetical":
syllables always can sound the same.
As the answer of Bob implies below, the recognition
for atypical people can be a critical path vital
issue. I would encourage you to use any means (four
months can be a very short period even for a developer
from India or Russia) to try to alleviate this kind of
people urgent necessity. Their problems are not very
different as that of the diabetic patients. Only they
and their families know about it; and, certainly they
would not object if you win a nobel prize or any other
prize, even with discussable tools as that of the
reverse engineering software.
Gerardo García
Madrid
--- spk2wrt@phoenix.edc.org wrote:
> [Moderator's note. Asim, speech recognition is
> certainly a valid
> dissertation project for an engineer, but I'm not
> sure exactly what aspect
> of SR you're interested in. As a confirmed
> non-engineer I don't really
> know how to advise you, but I do know there is LOTS
> to work on. One of the
> most interesting and difficult aspects of developing
> SR systems that I'm
> aware of is making systems that work for people with
> compromised,
> impaired, or in some way "non-standard" speech. I am
> also aware of work
> being done by speech scientists at MIT (and maybe
> elsewhere) developing SR
> models that include topical constraints as methods
> of improving
> recognition. Maybe others here have ideas? Good
> luck. Bob Follansbee]
>
> Dear,
>
> My name is Asim Shehzad. I am the old member of this
> group, but
> unfortunately I don't have lot of interaction on
> this group. I am doing
> software engineering from UK. For my dissertation
> project, I am interested
> in doing speech recognition systems. I want to get
> information about this topic. There is a lot of
> information on the group,
> but the people who are really in the field they know
> much better.
>
> First of all, is this good project? Second thing is
> is it possible to do
> it 4 months?
>
> Looking forward to hearing from you,
>
> Kind regards,
>
> Asim Shehzad
>
__________________________________________________
Do You Yahoo!?
Tired of spam? Yahoo! Mail has the best spam protection around
http://mail.yahoo.com
____________________________________________________________
To post a message to the spk2wrt list, send your message to:
spk2wrt@mail.edc.org. To reply to a message, simply use the reply button
in
your email program (do not change the subject line in any way). To access
the
spk2wrt archive containing a full list of all discussion threads to date,
point
your Web browser to: http://www.edc.org/spk2wrt/hypermail/. For questions
about
this list, please send a message to: spk2wrt@edc.org. For more
information
about the Speaking to Write project, please visit our Web site at:
http://www.edc.org/spk2wrt
____________________________________________________________
To post a message to the spk2wrt list, send your message to:
spk2wrt@mail.edc.org. To reply to a message, simply use the reply button in
your email program (do not change the subject line in any way). To access the
spk2wrt archive containing a full list of all discussion threads to date, point
your Web browser to: http://www.edc.org/spk2wrt/hypermail/. For questions about
this list, please send a message to: spk2wrt@edc.org. For more information
about the Speaking to Write project, please visit our Web site at:
http://www.edc.org/spk2wrt
This archive was generated by hypermail 2.1.2 : Tue Feb 08 2005 - 09:28:16 EST