The Maintenance and Tuning Phase

A speech application is unique in its need for tuning. The key is to capture authentic data from callers, and then use the data to refine — in effect to tune — the grammars, parameters, and other elements of the dialogue. The reason this must be done after deployment is that only authentic caller data are truly representative of the target population.

Some development organizations are nervous about “trying things out on real customers.” Managing the tuning phase is therefore critical. Stakeholders must understand that this is not “experimenting” on callers. Rather, this should be considered part of commissioning a new system. The system must be checked out and fine-tuned after deployment. The process is analogous to commissioning a new sailing vessel or nuclear reactor. The system is complex, and the tuning process is required to shake it down and stabilize it.

This phase might include any of the activities in the following list.

• Monitor logged calls and reports:
  • Establish and test tuning parameters
  • Use real caller data to refine prompts, grammars, and dialogues
  • Incorporate tuning suggestions and methodology in the system documentation
• Identify problem areas:
  • Grammars
  • Prompts
  • Parameter settings
  • Unexpected dialogue events
• Fix bugs
• Update the system with a controlled tuning release

To	See
Start from the beginning.	The Speech Project Lifecycle
Get more information on gathering the information to create a Design Requirements Specification.	The Planning and Discovery Phase
Get more information on creating an approved project plan.	The Design Phase
Get more information on building the speech application.	The Development Phase
Get more information on deploying the speech application.	The Deployment Phase