17 Tuning and deploying voice systems

 

This chapter covers

  • Defining voice system tuning concepts, including purpose and timelines
  • Understanding why some tuning is always necessary
  • Exploring how to analyze voice data and tune based on observations
  • Learn about tuning when access is limited on a development platform
  • Understanding how to avoid pitfalls before and after deployment

Tracking and tuning voice system performance is an iterative and ongoing process that starts during one or several pilot tests after you have worked through integration and functional testing. Limited tuning will be part of your testing during end-to-end and acceptance testing. Full-on tuning is only relevant once you have real user data from a fully tested and integrated system. You’ll understand why as you read through this chapter.

17.1  Tuning: what is it and why do you do it?

Tuning refers to the process of collecting and analyzing data from a live in-production voice system utilized by "real" end users, proposing solutions and implementing them. If you like solving puzzles you’ll love tuning. Let’s start with some definitions and reasons why you need to tune every voice system to some degree.

17.1.1    Why recognition accuracy isn’t enough

17.1.2    Analyzing causes of poor system performance

17.2  Tuning types and approaches

17.2.1    Log-based versus transcription-based tuning

17.2.2    Coverage tuning

17.2.3    Recognition accuracy tuning

17.2.4    Finding and using recognition accuracy data

17.2.5    Task completion tuning

17.2.6    Dialog tuning

17.2.7    How to prioritize your tuning efforts

17.3  Mapping observations to the right remedy

17.3.1    Reporting and using tuning results

17.4  How to maximize deployment success

17.4.1    Know when to tune

17.4.2    Understand tuning complexities to avoid pitfalls

17.5  What’s next?

17.6  Summary