Developing Speech Grammars: Speak Freely

SpeakFreely technology delivers accurate recognition results from natural language, open-ended prompts. This course explains the SpeakFreely technology for Nuance Recognizer 9 and describes situations where a SpeakFreely grammar offers the best solution for directing caller inquiries. The course walks through the development of a SpeakFreely grammar, including designing an open-ended question, collecting and labeling data, creating training files, testing and tuning statistical language models, and statistical semantic models, and generating a confidence engine.

This course is a three-session web-based course with homework. Please contact Nuance Speech University for details.

• Describe SpeakFreely’s role in a speech application
• List types of applications that can use SpeakFreely
• Explain the components of a SpeakFreely grammar
• Describe the SpeakFreely data collection process
• Explain the role and importance of labeling
• Prepare data for SpeakFreely
• Build SpeakFreely training Files
• Write a SpeakFreely wrapper grammar
• Train a SpeakFreely confidence engine
• Test a SpeakFreely grammar using Nuance Recognizer Tools

• Grammar Writers
• Speech scientists

1. SpeakFreely Overview
   • What is SpeakFreely?
   • When to use SpeakFreely
   • Components of SpeakFreely
2. Technology Review
   • XML Grammar Syntax
   • ECMAScript
3. SpeakFreely Data Collection
   • Determining Categories
   • Designing the Question
   • Collecting Data
4. Data labeling Process
   • Transcribing the Data
   • Sanitizing the Data
   • Labeling Data
5. Generating the SpeakFreely Grammar
   • The Training-Data Files
   • Using Subgrammars
   • Wrapper Grammar
   • Confidence Engine
6. Testing and Tuning
   • Accuracy in Automatic Labeling
   • Accuracy in Speech Recognition

• Introduction to Speech Recognition
• XML Grammar Fundamentals