📚 node [[voice options for your chatbot]]

Voice Options for your Chatbot

See [[Creating a Watson Chatbot with Discovery]], [[Integrating Discovery and Assistant]], and the [[Main AI Page]].

Natural interfaces vs Explicit interfaces

Speech uses less effort on behalf of the user, apparently. This might have taken a step backwards as early implementations have given users an expectation of a convoluted 'dialect' one must use when talking with machines.

Three main paths to Voice Assistants

  1. You can make the chatbot available through phonecalls while implementing the Assistant-to-Voice solution yourself (platform like Twilio).
  2. Implementing an application that marries Watson Assistant and Watson Speech-to-Text and Text-to-Speech APIs
  3. Phone-based systems using the IBM Voice Agent with Watson

Roll-your-own

Pros:

  • Easily accessible
    • Can be used over traditional phone numbers
  • No downloads required
    • No need to have the internet or install an app

Cons:

  • From-scratch implementation
    • You need to program and test the system yourself
  • Low flexibility

Watson Assistant-Text-to-Speech-to-Text

Pros:

  • Highly customisable and Flexible
    • If you need special functionality that no other system provides

Cons:

  • User usually has to download an application
  • Service orchestration is from scratch (IBM/etc infrastructure)
    • You might need to develop it for multiple OSs (iOS/Android)

Voice Agent with Watson

Pros:

  • Easily accessible
    • You're given your own SIP trunk
    • People can barge in and interrupt the assistant
    • Easier to go straight to a human agent
  • No downloads required
  • Service orchestration pre-done
    • Basically, the second option but handled for you

Cons

  • On-premises IBM Voice Gateway required for some advanced features.
    • Requires IBM's hybrid cloud-baremetal service
  • Intermediate flexibility
📖 stoas
⥱ context