Video Transcription

Asterisk Speech Recognition 101, part 2

Welcome to the second part of Asterisk Speech Recognition 101. We're going to continue talking about building your first speech app on Asterisk. First, let's talk about some of the different types of speech recognition applications that you normally build. There are two categories:

Natural Language Applications: Ask open-ended questions and allow for a large range of responses.

• Very natural and easy to use.
• Have lower recognition accuracy and higher development time. This is because as you increase the size of your grammar, generally your accuracy decreases.

Directed Dialogue Applications: Use prompts that guide the user to give specific responses.

• Use prompts that guide the user to give specific responses.
• Cover fewer possible responses than natural language, but often have better success rates.
• Have lower design and development time compared to natural language.

Prompts

• Avoid "say" menus.
• Give the user a mental model of the application with each prompt.
• Hire voice talent when possible to make the app more personal and engaging. Avoid TTS except where necessary.
• Include confirmation prompts.

Speech Recognition on Asterisk

A generic speech API (res_speech) was introduced in 1.4. A patch exists for 1.2. This provides a number of dial plan applications related to speech (app_speech_utils). These are also available via AGI.

Connector Bridge

To tie a speech recognition engine into the Asterisk speech API, a second module is needed: the Connector Bridge. The connector for the LumenVox Speech Engine is res_speech_lumenvox.so. It was developed and is maintained by Digium.

In Part 3, we'll talk more about what grammars are and how to write them.