[Olympus developers 447]: Re: how to enable multiple users simultaneously.

Aasish Pappu aasishp at gmail.com
Sat Feb 8 14:44:19 EST 2014


For a telephone interface, you could use a softswitch like freeswitch that
supports multiple sessions (default is 1000) and protocols to communicate
with an application.

We have tried to provide a preliminary interface between the audioserver
component and the freeswitch server. There's a tarball that you can start
from. The freeswitch application writes incoming audio data to a named-pipe
for every new session. This audio data is read by an instance of
audioserver component and forwarded to the recognizer (pocketsphinx).

http://www.speech.cs.cmu.edu/apappu/pubdl/fs-dialog-interface.tar.gz

There is an example application (with example configuration) in the
tarball. It also includes customized audioserver and the pocketsphinx
engine. If you want to rest of the olympus components, you have to figure
out how they will interact with the freeswitch application.




On Sat, Feb 8, 2014 at 1:39 PM, Nitin Dhawan
<nitin at imarketingadvantage.com>wrote:

> Thanks. This does look a little complicated. We will try to understand
> this better and come back for further clarifications if required.
>
>
>
> I do have another question on how this can be done using telephone
> interface. I will put it in another thread.
>
>
>
>
>
> *From:* Aasish Pappu [mailto:aasishp at gmail.com]
> *Sent:* Saturday, February 8, 2014 3:21 AM
> *To:* Nitin Dhawan
> *Cc:* olympus-developers
> *Subject:* Re: [Olympus developers 443]: how to enable multiple users
> simultaneously.
>
>
>
> The architecture natively doesn't support multiple users (I assume you
> meant simultaneous sessions).  However, there were instances where you
> could run multiple instances of some of the components (Audio Server, DM,
> InteractionManager, Hub) corresponding to each session. Since those
> components are session specific. There are few components (ASR, NLU, NLG)
> which can run as services i.e., doesn't have to be instantiated. All of
> this
>
> customization will require understanding of how the components interact.
>
>
>
> The short answer is there is no easy recipe to customize the architecture
> for multiple sessions.
>
>
>
> There are few example systems that were built to handle certain aspects of
> the simultaneous sessions.
>
>
>
> 1. Multiparty Interaction: System interacting with more than user at a
> time in a single interactive session.
>
>
>
> repo: http://trac.speech.cs.cmu.edu/repos/olympus/branches/MultiOly1.0/
>
>
>
> We instantiate dialog manager, hub for each user to keep track of their
> slot/value pairs (if different from other user). We don't have to
> instantiate the IM and AudioServer in this case because both users are
> listening on same channel (loud speakers) and talking on same channel
> (kinect microphone array).  While talking to the users, we don't want the
> dialog manager to request for multiple nlg prompts for each user, instead
> we synchronize the system's response from DM to one of the users or both
> users. We do this using a new component called conversation manager that
> talks to interaction-manager on behalf of both the dialog managers. More
> details on this work in
> http://link.springer.com/chapter/10.1007/978-3-642-39330-3_12
>
>
>
> 2. Human robot teams: Single user interacting with multiple robots.
>
>
>
> We instantiate dialog manager, hub, robot backend for each robot. The
> dialog context for each robot is independent from each other. Here each
> robot's DM responds to the user based on a trigger phrase (e.g., RobotA
> listen, RobotB listen).
>
>
>
> repo: http://trac.speech.cs.cmu.edu/repos/teamtalk/branches/tt-olympus2/
>
>
>
>
>
>
>
>
>
> On Fri, Feb 7, 2014 at 3:44 PM, Nitin Dhawan <
> nitin at imarketingadvantage.com> wrote:
>
> Hi
>
>
>
> This might be a stupid question, but how to enable multiple users to
> access the system at the same time. How does the system have to be deployed.
>
>
>
> Based on all the documentation we have read so far, one can start all the
> servers and get either a speech , or tty r java-tty interfact where one
> user can interact at a time.
>
>
>
> We also want to enable authentication of this system, using a single sign
> on with another system. But thats really the next step.
>
>
>
> Thanks
>
> Nitin Dhawan
>
>
>
>
>
> --
> Aasish Pappu
>



-- 
Aasish Pappu
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.srv.cs.cmu.edu/pipermail/olympus-developers/attachments/20140208/6d8c47a5/attachment-0001.html>


More information about the Olympus-developers mailing list