Previous Table of Contents Next


The following description of the development process assumes that the scope of the project is known and a preliminary specification has been developed. This specification would include:

Exhibit 1. Development Task Sequence for ASR Applications

Task Description

1. Requirements Definition Define dialog, APIs, text plan, load characteristics, measurement methods, and preliminary rollout plan.
2. Standalone Prototype Iteratively develop and test standalone prototype and API, Development install vocabulary, tune for regional accents.
3. Integration Integrate with IVR and test (or test with test databases), simulate load testing.
4. Pilot Test Test functional operation with a limited user set, modify as necessary.
5. Detailed Rollout Plan Develop detailed rollout plan, install equipment for rollout.
6. Rollout Full system test followed by implementation.

  A description of the proposed application.
  A description of the operating environment, including existing hardware and software systems that will interface with the ASR application or any preference for hardware and software that imposes a constraint on the ASR solution.
  A preliminary description of the anticipated dialog between users and the system.
  A list of the specialized vocabulary words or phrases needed by the application.

Because speech recognition applications often interface with or support mission- critical systems, a cautious six-step approach that includes extensive testing before implementation is required. Exhibit 1 summarizes the tasks involved in the approach.

Task 1: Requirements Definition. The development team defines the application requirements in sufficient detail to create a more specific project plan and allocate assignments and target schedules. This documented definition includes:

  A definition of the dialog to be conducted between the user of the speech recognition system and the ASR system. This dialog anticipates such system behavior as points of dropout to a live operator, alternate use of touchtone input, and transfers to other voice routines.
  A definition of the application program interfaces (APIs) to cooperating voice response units or other equipment or software.
  A definition of message formats and message-passing protocols to support any required interfaces, accesses, or updates to production databases. To protect the integrity of the company’s databases, the ASR system should not directly access them in any manner except through a defined procedure/protocol provided by the technical staff responsible for the database. The interface protocol permits sending and receiving messages between the voice recognition system and the database access programs.
  Test plans for API testing, controlled environment testing, pilot testing, and acceptance testing.
  Identification and description of a pilot test group.
  Identification and description of anticipated load characteristics.
  A definition of performance measurement parameters for evaluating the performance of the system during pilot test and under full operation.
  A preliminary plan for rollout of the application to full use.

During task 1, the company should also acquire and install a workstation-based development environment for the creation and testing of the prototype and for use in future development and maintenance. The ASR application development toolkit is installed on the workstation.

Upon completion of task 1, the requester’s representatives review and approve the resulting definitions and plans.

Task 2: Prototype Development. The team employs a technique of incremental prototyping to develop and iteratively test a prototype application in a standalone environment. Technical personnel assigned to the project team are trained in the use of the toolkit by the ASR system provider during this task. This basic training in use of the toolkit ensures that the staff becomes familiar with the tools and methods used in voice recognition applications development.

The team installs the vocabulary and any variations that are anticipated to handle regional accents, creates the dialog, implements word spotting if required, implements accuracy performance measurement routines, and, in general, establishes a complete prototype of the application in a standalone environment. Iterative development ensures that the prototype has continually been tested as it is developed.

At the conclusion of this task, the operating prototype should satisfy the functional requirements of the voice recognition portion of the application. The operating prototype is essentially the runtime version of the voice recognition portion of the application, ready for integration into the overall application environment.


Previous Table of Contents Next

Copyright © CRC Press LLC