We previously wrote extensively about Alexa. How to get started, understand the voice interaction model, low-code development environment for Alexa skills. Use the right panel search bar to explore these articles. Today we want to expand the talk to be more general and include other voice assistant technology such as wit.ai acquired by Facebook.
Voice assistants have evolved to be modern, and AI enabled in 2022. Now they are smart voice assistants. They are not routes - rule based logic that routes users to the right department, phone book, voicemail. They are intelligent, often enabled by natural language processing (NLP) capabilities.
Some keywords have not changed:
Intent: what is the action that user wants to achieve, e.g. buy eggs
Entities: important nouns in utterances, which helps us understand the user's intent and the subject and object the intent is acting on. e.g. "egg". Can also be geographic locations, time, dates etc. Select a part of utterance as wit location entity example facebook wit demo distance between Seattle and Menlo Park
Utterance: what the user actually tells the voice assistance e.g. "Put some organic eggs in my shopping cart." Utterances can come in many forms. Must be thoroughly tested include many examples, and be natural.
Be sure to always handle errors: there should be a fall back answer "I am sorry I cannot understand what you said. Try say xyz. You can also contact us at xyz"
Twilio now also offers intelligent AI based intent classification. Wit.ai can be used with cloud functions and twilio studio. Wit.ai gives user NLP capabilities without code. Providing utterance examples and labeling entities is easy. Click and submit. Wit.ai also allows easy incorporation of smart replies in Facebook Messenger. It's useful for business pages as well. I might be biased because I participated in wit.ai hackathons and was impressed!
At the hackathon, I was able to test out wit.ai using curl
curl -i -X GET "https://graph.facebook.com/{page-id}?fields=about,attire,bio,location,parking,hours,emails,website&access_token={page_access_token}"
Be sure to replace page-id and page_access_token with your own parameters.
I was able to hook wit.ai API with twilio workflow. What's even cooler is that you can use postman to test wit.ai API! The facebook team used glitch to demo their app with node.js, webhook . I liked the idea of the glitch but it had a learning curve for me, so I quickly gave up. ,
The workflow of wit.ai is :
- Create a facebook developer app and get a developer API
- create a facebook page to use wit.ai in messenger
Messenger apps are associated with a specific Facebook Page.
Using wit.ai on Facebook page can help seller answer questions, automatically make product recommendations. Using Facebook tools has one caveat the license is often not fully open sourced. Your custom app likely is not private to Facebook, just like source code of iphone apps aren't private. Apple can potentially make clones. Facebook has similar terms too. Watch out.
Alexa
When we first got started with Alexa there were only 25000 skills. How many skills does Alexa have today? Comment below.
Easy alexa quiz app: true or false questions, question banks. Give hints when users need it.
No code alexa? Program alexa skills without coding? There was a service called Storyline. Alexa seems to have its own inhouse AWS solution called Blueprint, perhaps the easiest way to make an Alexa app without coding.
Additional Twilio offerings:
Twilio Machine Learning
No comments:
Post a Comment