Concurrent engines running on the same client

Is it possible to have a single client receiving messages from 2 SAPI
engines running at the same time?

That is, the default SAPI recognizer works fine for command and
control and most of my dictation cases. But, I now have access to
another engine that can be customized with a different language model.
I would want to create a session with that engine (while giving it an
ARPA compliant language model to customize it). For example, I would
want to create a language model for the music on a computer. That
language model could hold entries like "play the song candle in the
wind", or "play music by Elton John". If both engines are running
at
the same time (and both could have dictation enabled at the same
time), I would expect my music enabled vocabulary to return the result
if I say music related content, and straight dictation otherwise.

Is that possible?

Hi Philippe,

One of my colleagues provided this response:

In this case they don't want two engines running.  If you have two engines 
running dictation grammars you will generally get recognitions from both 
engines for every utterance.  Then you have no practical and reliable way of 
knowing which to pick.

Instead, they simply want both grammars active in a single engine, so that 
engine can pick which recognition to return.

For their example below, they would create a CFG, not an arpa LM.  If they 
want the user to have more flexibility in what is said then they could 
indeed use something like an ARPA LM.  But then they will need a subsequent 
parse stage to figure out the semantics so they know what the user wants to 
do.  Given the examples below, I believe it would be significantly easier 
for them to achieve their goals with a properly constructed CFG than with an 
SLM.

One thing that isn't clear to me from the message is if they have some third 
party engine that is allowing them to have an ARPA LM, or if they think they 
can do this with our tools or if they are aware of the DRK and are asking 
about that.

In Vista we do have multiple dictation models active at the same time.  The 
other LMs have prefixes, for example you can say "How do I" followed
by 
something from the HowDoI LM (that is used to look things up in help) or 
"insert URL" followed by something from the URL LM or
"spell" followed by 
something from the spelling LM.

-Lesley
-- 
This email is provided "AS IS" with no warranties, and confers no
rights




"Philippe Roy" <speechvibe@googlemail.com> wrote in message 
news:9d94b1a3-2f15-4484-84a5-38b0db65c386@41g2000hsy.googlegroups.com...
> Is it possible to have a single client receiving messages from 2 SAPI
> engines running at the same time?
>
> That is, the default SAPI recognizer works fine for command and
> control and most of my dictation cases. But, I now have access to
> another engine that can be customized with a different language model.
> I would want to create a session with that engine (while giving it an
> ARPA compliant language model to customize it). For example, I would
> want to create a language model for the music on a computer. That
> language model could hold entries like "play the song candle in the
> wind", or "play music by Elton John". If both engines are
running at
> the same time (and both could have dictation enabled at the same
> time), I would expect my music enabled vocabulary to return the result
> if I say music related content, and straight dictation otherwise.
>
> Is that possible?
>
> Thank you 

One more comment:

One other thing: depending on the flexibility they want, they should look at 
the subset feature in SAPI 5.3  Unfortunately, it doesn't appear to be 
documented in MSDN. A bug has been opened on that.

Regards,
Lesley

-- 
 This email is provided "AS IS" with no warranties, and confers no
rights

"Lesley Carmichael [MSFT]"
<lesley.carmichael@online.microsoft.com> wrote in 
message news:%23HRr8nKWIHA.5980@TK2MSFTNGP04.phx.gbl...
> Hi Philippe,
>
> One of my colleagues provided this response:
>
> In this case they don't want two engines running.  If you have two engines

> running dictation grammars you will generally get recognitions from both 
> engines for every utterance.  Then you have no practical and reliable way 
> of knowing which to pick.
>
> Instead, they simply want both grammars active in a single engine, so that

> engine can pick which recognition to return.
>
> For their example below, they would create a CFG, not an arpa LM.  If they

> want the user to have more flexibility in what is said then they could 
> indeed use something like an ARPA LM.  But then they will need a 
> subsequent parse stage to figure out the semantics so they know what the 
> user wants to do.  Given the examples below, I believe it would be 
> significantly easier for them to achieve their goals with a properly 
> constructed CFG than with an SLM.
>
> One thing that isn't clear to me from the message is if they have some 
> third party engine that is allowing them to have an ARPA LM, or if they 
> think they can do this with our tools or if they are aware of the DRK and 
> are asking about that.
>
> In Vista we do have multiple dictation models active at the same time. 
> The other LMs have prefixes, for example you can say "How do I"
followed 
> by something from the HowDoI LM (that is used to look things up in help) 
> or "insert URL" followed by something from the URL LM or
"spell" followed 
> by something from the spelling LM.
>
> -Lesley
> -- 
> This email is provided "AS IS" with no warranties, and confers no
rights
>
>
>
>
> "Philippe Roy" <speechvibe@googlemail.com> wrote in message

> news:9d94b1a3-2f15-4484-84a5-38b0db65c386@41g2000hsy.googlegroups.com...
>> Is it possible to have a single client receiving messages from 2 SAPI
>> engines running at the same time?
>>
>> That is, the default SAPI recognizer works fine for command and
>> control and most of my dictation cases. But, I now have access to
>> another engine that can be customized with a different language model.
>> I would want to create a session with that engine (while giving it an
>> ARPA compliant language model to customize it). For example, I would
>> want to create a language model for the music on a computer. That
>> language model could hold entries like "play the song candle in
the
>> wind", or "play music by Elton John". If both engines
are running at
>> the same time (and both could have dictation enabled at the same
>> time), I would expect my music enabled vocabulary to return the result
>> if I say music related content, and straight dictation otherwise.
>>
>> Is that possible?
>>
>> Thank you
>