A
fully configured FGC/SRS System is similar to a three-stage
client/server system. The client executes a speech recognition
request and sends it over the network to a Load Balancing
Server (LBS) within the FGC/SRS System. The LBS passes
the request to a Host Server (HS). The HS chooses one
of several FGC/MP-7 parallel processors to perform the
actual recognition task and when completed, to transmit
the result back to the client.
The
FGC/SRS System has several layers of redundancy:
-
The Load Balancing Server insulates the client program
from the multiple FGC/MP-7 parallel processors and,
-
The FGC/MP-7 Host insulates the client program from
any of the multiple PE recognizers that might be off
line.
Communication
between layers uses peer-to-peer TCP/IP protocol. The
messaging protocol between the Application and Load
Balancing Server is identical to that between the Load
Balancing Server and the Host Server. This in turn is
identical to the protocol between the Host Server and
the PE server.
SYSTEM
HARDWARE
FGC/MP-7 Recognition Engine
Consists
of seven Pentium processors
FGC/MP-7 Host Computer
Pentium
processor with cache and SDRAM
Four
Gigabyte hard drive capacity
Redundant power supplies throughout
Each
includes expandable high performance memory
Network
interface 10 or 100 Mbit/sec Ethernet
Packaged
to simplify client/server network integration
SOFTWARE
ARCHITECTURE:
The
primary components are:
-
Client (Application) Program
-
The Load Balancing Server Program
-
The PE Server Program
They
share many common elements.
Client
(Application) Program: The client
program communicates with the Load Balancing Server
Program using the Client Library Module.
This Module uses TCP/IP protocols to communicate with
the Load Balancing Program
Load
Balancing Server: Communication with
the Application Program (getting requests and returning
results) is serviced by the Network Handler
Module. It places incoming network requests onto
a Request Queue. Processed results (from within
the server) are taken off the Result Queue and written
to the network.
The
Request Handler Module
moves requests and results between the two queues and
the Load Balancing Service Module, which actually executes
each request. The Load Balancing Service Module sends
requests to the next available Host Server using the
Client Library Module.
PE
Server: The PE Server does the actual
speech recognition. Requests are processed by the Recognition
Service Module. The Recognition Service Module uses
the Speech Library Module to perform Speech Recognition
functions.
SYSTEM
SOFTWARE
FGC/MP-7 Administrative and Diagnostic Utility
FGC/MP-7
Application Program Interface Library (API)
FGC/SRS
Speech Library v1.0 (Speech Recognition Engine)
Client
Library
