Archived from groups: comp.sys.palmtops.pilot (More info?)

Just reading some archive posts on this subject.
Says it can't be done, but that was 4 years ago.

Is there any good speech to text recognition software
for the palm os? I use Sony Clie 80V.

What would be ideal would be something like the
following:

1. Reasonably accurate conversion to text.
2. Reasonably fast conversion. I do not want to
pause a 1/2 second between words.
3. It goes without saying, but a vocabulary larger
than mine. (Not that I'm a scholar.)
4. Can be used in any applications where graffiti
input is used.
5. Can be mixed with graffiti.
6. Has a way of distinguishing between punctuation
commands like 'period'. And when I want to
insert the word 'period'.
7. Has a 'record' button, so it knows when I'm
talking to myself, and when I'm talking to
my pda.

Anyways, is there anything out there like this?

thanks.

Archived from groups: comp.sys.palmtops.pilot (More info?)

Love Rhino wrote:

> Is there any good speech to text recognition software
> for the palm os? ...
>
> 1. Reasonably accurate conversion to text.
> 2. Reasonably fast conversion. I do not want to
> pause a 1/2 second between words.
> 3. It goes without saying, but a vocabulary larger
> than mine. (Not that I'm a scholar.)
> 4. Can be used in any applications where graffiti
> input is used.
> 5. Can be mixed with graffiti.
> 6. Has a way of distinguishing between punctuation
> commands like 'period'. And when I want to
> insert the word 'period'.
> 7. Has a 'record' button, so it knows when I'm
> talking to myself, and when I'm talking to
> my pda.

I do not know everything on the subject, but I would be TREMENDOUSLY
surprised if the answer were yes. I have been using SR software for 5
years now as an alternative to transcriptionist in my medical practice.
The hardware requirements for SR are so demanding that it would be
hard to imagine a PDA/Handheld doing SR. There is now only a single
software product left in the SR marketplace, Dragon Naturally Speaking
(unless you count the Mac-platform-only iListen). Speech recognition in
real-time (within moments of being uttered) requires incredible
computing power. It is one of the few applications which actually does
take advantage of the power of computers these days. Dragon's current
version, and even the previous several versions, all require about
800-1000 mhz P3/P4, 512 RAM, Win2kPro or better, and don't forget a
superior noise cancelling mic with excellent accoustics. The little
voice-memo mic on your Clie (or my T3) is laughably inadequate.

Given these minimal requirements, I don't see how any PDA on the market
today can do SR. But I'd love to hear a different answer or new
development.

Best regards,
-- Josh Steinberg

Archived from groups: comp.sys.palmtops.pilot (More info?)

"Joshua Steinberg" <jsteinb1@twcny.rr.com> wrote in message
news:aMdNc.56215$KU.28334@animal.nntpserver.com...
> Love Rhino wrote:
>
> > Is there any good speech to text recognition software
> > for the palm os? ...
> >
> > 1. Reasonably accurate conversion to text.
> > 2. Reasonably fast conversion. I do not want to
> > pause a 1/2 second between words.
> > 3. It goes without saying, but a vocabulary larger
> > than mine. (Not that I'm a scholar.)
> > 4. Can be used in any applications where graffiti
> > input is used.
> > 5. Can be mixed with graffiti.
> > 6. Has a way of distinguishing between punctuation
> > commands like 'period'. And when I want to
> > insert the word 'period'.
> > 7. Has a 'record' button, so it knows when I'm
> > talking to myself, and when I'm talking to
> > my pda.
>
> I do not know everything on the subject, but I would be TREMENDOUSLY
> surprised if the answer were yes. I have been using SR software for 5
> years now as an alternative to transcriptionist in my medical practice.
> The hardware requirements for SR are so demanding that it would be
> hard to imagine a PDA/Handheld doing SR. There is now only a single
> software product left in the SR marketplace, Dragon Naturally Speaking
> (unless you count the Mac-platform-only iListen). Speech recognition in
> real-time (within moments of being uttered) requires incredible
> computing power. It is one of the few applications which actually does
> take advantage of the power of computers these days. Dragon's current
> version, and even the previous several versions, all require about
> 800-1000 mhz P3/P4, 512 RAM, Win2kPro or better, and don't forget a
> superior noise cancelling mic with excellent accoustics. The little
> voice-memo mic on your Clie (or my T3) is laughably inadequate.
>
> Given these minimal requirements, I don't see how any PDA on the market
> today can do SR. But I'd love to hear a different answer or new
> development.
>
> Best regards,
> -- Josh Steinberg

The other side of this viewpoint is that very limited SR is possible on a PDA, after
all my cell phone, which is two years old now, has it for dialing.

Adam

Archived from groups: comp.sys.palmtops.pilot (More info?)

Previously on comp.sys.palmtops.pilot, Adam Helberg said:

; The other side of this viewpoint is that very limited SR is possible on a PDA, after
; all my cell phone, which is two years old now, has it for dialing.

So does mine. It does it by recording specific phonetic sequences.
And about 1/4 of the time, it is unable to distinguish between three
different voice names that sound sufficiently different to my ear.

--
Jeffrey Kaplan www.gordol.org
The from userid is killfiled Send personal mail to gordol

"Civilization has supplanted natural selection through legislation and
medical technology. We now have people in positions of responsibility
and authority who are so stupid that two hundred years ago they would
have been eaten by something on their way to the outhouse." - 'Poe's
Theorem of Evolution'

Archived from groups: comp.sys.palmtops.pilot (More info?)

Jeffrey Kaplan <cspp@gordol.org> wrote in
news:9c5bg09ol5p877516kn02q4hjdfmpf05mo@news20.forteinc.com:

> Previously on comp.sys.palmtops.pilot, Adam Helberg said:
>
> ; The other side of this viewpoint is that very limited SR is possible
> on a PDA, after ; all my cell phone, which is two years old now, has
> it for dialing.
>
> So does mine. It does it by recording specific phonetic sequences.
> And about 1/4 of the time, it is unable to distinguish between three
> different voice names that sound sufficiently different to my ear.
>

Did you mean to say....



-S-

Archived from groups: comp.sys.palmtops.pilot (More info?)

On Mon, 26 Jul 2004 22:52:27 +0000, Adam Helberg wrote:

> The other side of this viewpoint is that very limited SR is possible on
> a PDA, after all my cell phone, which is two years old now, has it for
> dialing.

The cell phone doesn't to Speech-to-Text, it merely tries to match a
spoken phrase to a recording. It does so without "knowing" what either
phrase "is", a requisite for Speech-to-Text.

--
Lenroc

Archived from groups: comp.sys.palmtops.pilot (More info?)

All:

Simply put, this is currently technologically not feasible. There is
currently no PDA available that has either the processing power or
storage capacity to do true, phrase-based speech-to-text. Some
companies have done some experiments with taking a sample and sending
it wirelessly to a server, where the user's speech is converted to
text and then transmitted back to the PDA, but the problem here is the
bandwidth required to send a high quality sample (16khz or better)
doesn't exist either.

Apple's iPod has the storage capacity, but unfortunatetly only samples
at 8khz (again, not high enough for speech recognition).

The only way we have found to do "remote" speech recognition is to
record your voice on a high-quality device that does not compress or
clip the recordings, such as the Olympus Digital Voice Recorders, and
then transcribe the audio files later on the computer. This is the
method we use in our iListen Transcription Solution.

Chuck Rogers, Chief Evangelist
MacSpeech, Inc.


"Adam Helberg" <sendspamhere@yahoo.com> wrote in message news:<LWfNc.17334$Qu5.12382@newsread2.news.pas.earthlink.net>...
> "Joshua Steinberg" <jsteinb1@twcny.rr.com> wrote in message
> news:aMdNc.56215$KU.28334@animal.nntpserver.com...
> > Love Rhino wrote:
> >
> > > Is there any good speech to text recognition software
> > > for the palm os? ...
> > >
> > > 1. Reasonably accurate conversion to text.
> > > 2. Reasonably fast conversion. I do not want to
> > > pause a 1/2 second between words.
> > > 3. It goes without saying, but a vocabulary larger
> > > than mine. (Not that I'm a scholar.)
> > > 4. Can be used in any applications where graffiti
> > > input is used.
> > > 5. Can be mixed with graffiti.
> > > 6. Has a way of distinguishing between punctuation
> > > commands like 'period'. And when I want to
> > > insert the word 'period'.
> > > 7. Has a 'record' button, so it knows when I'm
> > > talking to myself, and when I'm talking to
> > > my pda.
> >
> > I do not know everything on the subject, but I would be TREMENDOUSLY
> > surprised if the answer were yes. I have been using SR software for 5
> > years now as an alternative to transcriptionist in my medical practice.
> > The hardware requirements for SR are so demanding that it would be
> > hard to imagine a PDA/Handheld doing SR. There is now only a single
> > software product left in the SR marketplace, Dragon Naturally Speaking
> > (unless you count the Mac-platform-only iListen). Speech recognition in
> > real-time (within moments of being uttered) requires incredible
> > computing power. It is one of the few applications which actually does
> > take advantage of the power of computers these days. Dragon's current
> > version, and even the previous several versions, all require about
> > 800-1000 mhz P3/P4, 512 RAM, Win2kPro or better, and don't forget a
> > superior noise cancelling mic with excellent accoustics. The little
> > voice-memo mic on your Clie (or my T3) is laughably inadequate.
> >
> > Given these minimal requirements, I don't see how any PDA on the market
> > today can do SR. But I'd love to hear a different answer or new
> > development.
> >
> > Best regards,
> > -- Josh Steinberg
>
> The other side of this viewpoint is that very limited SR is possible on a PDA, after
> all my cell phone, which is two years old now, has it for dialing.
>
> Adam

Archived from groups: comp.sys.palmtops.pilot (More info?)

On Mon, 26 Jul 2004 09:41:31 -0700, Love Rhino wrote:

> Just reading some archive posts on this subject.
> Says it can't be done, but that was 4 years ago.
>
> Is there any good speech to text recognition software
> for the palm os?

I, like others here, don't think it's possible.

However, you might be able to get something close by using a PDA with
"Voice Memo" functionality, then exporting that to a PC-based SR software
package.

This wouldn't be useful in the way you are asking about though
(controlling the PDA), but rather only (possibly) for transcription.

--
Lenroc

Archived from groups: comp.sys.palmtops.pilot (More info?)

"Adam Helberg" <sendspamhere@yahoo.com> wrote in message

snip snip

> The other side of this viewpoint is that very limited SR is possible on a PDA, after
> all my cell phone, which is two years old now, has it for dialing.
>
> Adam

Unfortunately, a lot of cell phone manufacturers call their voice dial
"SR." Some of the experts seem to hugely disagree: some say that cell phones
voice dialling is so constrained that it doesn't meet the threshold
definition for SR.

It's like that old argument about what is the minimum definition for "language."

I wonder, though, since dragon,L&H, ibm, etc. began selling their "SR" products
such a long time ago (I think I tried using them in the days of p5-60MHz),
are ARM-based Palms capable of more power (even on a 64Mb mostly stuffed with
palm apps and pre-cobalt OS technology)?

Likewise, when palm started using the dual core omax chips, one of the
"coming" benefits claimed was SR. Now that omax isn't part of palmOne's
plans and I haven't heard such claims, how well can future palms handle the
rigors of SR?

What's happening with Wince?

Archived from groups: comp.sys.palmtops.pilot (More info?)

As with all technology the rubber meets the road with the user and their
expectations. Marketeers can spin and technologists can quibble but the
real issue is will a user's reasonable expectations be met. As to free
form speech recognition the error rates on any platform are too high for
the technology to be useful for any but the disabled. Fixing mistakes in
a generated document gets old fast -- particularly since SR implies hands
free. That's not to say that some applications with limited vocabularies
and without the requirement for speaker Independence can't work well. For
example, commands or voice dialing of telephones. It all depends on the
application and the reasonable user expectations. IMHO of course.

On 26 Jul 2004 21:56:44 -0700, spamnot@notspam.com <dejpd@yahoo.com> wrote:

> "Adam Helberg" <sendspamhere@yahoo.com> wrote in message
>
> snip snip
>
>> The other side of this viewpoint is that very limited SR is possible on
>> a PDA, after
>> all my cell phone, which is two years old now, has it for dialing.
>>
>> Adam
>
> Unfortunately, a lot of cell phone manufacturers call their voice dial
> "SR." Some of the experts seem to hugely disagree: some say that cell
> phones
> voice dialling is so constrained that it doesn't meet the threshold
> definition for SR.
>
> It's like that old argument about what is the minimum definition for
> "language."
>
> I wonder, though, since dragon,L&H, ibm, etc. began selling their "SR"
> products
> such a long time ago (I think I tried using them in the days of
> p5-60MHz),
> are ARM-based Palms capable of more power (even on a 64Mb mostly stuffed
> with
> palm apps and pre-cobalt OS technology)?
>
> Likewise, when palm started using the dual core omax chips, one of the
> "coming" benefits claimed was SR. Now that omax isn't part of palmOne's
> plans and I haven't heard such claims, how well can future palms handle
> the
> rigors of SR?
>
> What's happening with Wince?



--
_____________
Thank a veteran

Archived from groups: comp.sys.palmtops.pilot (More info?)

"Jeffrey Kaplan" <cspp@gordol.org> wrote in message
news:9c5bg09ol5p877516kn02q4hjdfmpf05mo@news20.forteinc.com...
> Previously on comp.sys.palmtops.pilot, Adam Helberg said:
>
> ; The other side of this viewpoint is that very limited SR is possible on a PDA,
after
> ; all my cell phone, which is two years old now, has it for dialing.
>
> So does mine. It does it by recording specific phonetic sequences.
> And about 1/4 of the time, it is unable to distinguish between three
> different voice names that sound sufficiently different to my ear.
>
> --
> Jeffrey Kaplan www.gordol.org
> The from userid is killfiled Send personal mail to gordol
>

I try to use very different names for the numbers and my success rate is better than
that. It errs mainly if there is background noise.

In any case I recall an old SR program several years ago, when it had doubt about a
word it would display a list of words and you would select the word by saying the
number. So SR does not have to be perfect to be usable.

Adam

Archived from groups: comp.sys.palmtops.pilot (More info?)

On Tue, 27 Jul 2004 05:49:05 +0000, Adam Helberg wrote:

> In any case I recall an old SR program several years ago, when it had
> doubt about a word it would display a list of words and you would select
> the word by saying the number. So SR does not have to be perfect to be
> usable.

True, but that's not really the issue here.

The issue here is that to be considered "SR", the system must be
interpreting what it hears, not just matching it to prerecorded samples.

--
Lenroc

Archived from groups: comp.sys.palmtops.pilot (More info?)

"George Fragos" <fragos@earthlink.net> wrote:
> As with all technology the rubber meets the road with the user and their
> expectations. Marketeers can spin and technologists can quibble but the
> real issue is will a user's reasonable expectations be met.

This is certainly true. One problem with this technology that the marketeers
spun this way too hard and there was no way at the time that SR would come
close to those promises. However, IMO if the marketeers' promises had been
more muted then this is a technology that could then and now deliver some
advantage to some people some times.

> As to free
> form speech recognition the error rates on any platform are too high for
> the technology to be useful for any but the disabled.

There are people then and now who still swear by it and use it constantly.
(Ex: some doctors use it daily for transcription). Of course, this is
far from the jetsons-like future of promised ubiqitous SR. (Your refridgerator
greets you and says "Would you like your OJ this morning?" )

> Fixing mistakes in
> a generated document gets old fast -- particularly since SR implies hands
> free.

This is true. But for some needs, the benefits outweigh the downsides here.

<some comments deleted>

chuck.rogers@gmail.com (Chuck Rogers) wrote:
<some comments deleted>

> The only way we have found to do "remote" speech recognition is to
> record your voice on a high-quality device that does not compress or
> clip the recordings, such as the Olympus Digital Voice Recorders, and
> then transcribe the audio files later on the computer. This is the
> method we use in our iListen Transcription Solution.
>
> Chuck Rogers, Chief Evangelist
> MacSpeech, Inc.

How well does this work?

I've tried this approach (record now, process later) with varying results, both
via getting various dictaphones and buying the bundles. So far, disappointment.

I've tried microcassette recorders, high-quality microcassette recorders,
high-quality microcassette recorders with high-quality mics (grin), and
various digital recorders.

The error rate goes way up to the point of being not useful so far for me.
I have tried to control for backround noise, mic quality, other distractions
and so on.

1) Is it do-able with any microcassette recorder?
2) Do you need to add a secondary mic?
3) Is it do-able with any palm with a built-in mic (ex: tungsten t3 zire72,
NOT tungsten C).
4) Do pocket pcs do this any better? Worse?