Conversational Avatars

Cocohub offers a platform for building conversational avatars. I gave it a try and was very pleased with the results.

You start by choosing an avatar from a fairly wide selection of off-the-shelf body types. Next, you choose a Amazon Polly voice to match.

Once you’ve decided what your avatar will look and sound like, you connect and/or build the components that power the avatar’s personality and conversational abilities.

The Power of Prebuilt Components

One of Cocohub’s selling points is the library of prebuilt components it offers. These components cover common basic functions, such as asking for someone’s name, or getting their physical or email address. They also cover more complex functions, such as accessing a calendar, sending an email, or taking a survey.

Cocohub also offers integration with open source Blenderbot. This is a powerful tool enabling your avatar to carry on a conversation autonomously, without any pre-scripted dialogue.

The Sensorium virtual being I wrote about previously used a similar neural net powered language model to respond autonomously to a person’s input. It’s always a bit of a gamble using this technology, since you can never predict what the bot will say or if the content will be appropriate. But the fact that Cocohub offers it as an option puts another very useful tool into your toolbox.

Hosting Your Avatar on Multiple Platforms

Once you’ve built a conversational avatar on Cocohub, you can choose to host it on a website, or you can have it appear via interactive video. If you forego the avatar you can publish the conversational bot to a phone channel, and if you just have a text-based chatbot, you can launch it on Messenger, What’s App, and Telegram.

I really like the option to run the avatar through video. Once the video is published, the person interacting with your avatar just needs to provide access to their microphone and say the phrase “Wake up” to start the conversation.

So far Cocohub is the only platform I’ve seen that enables you to quickly build a fully conversational avatar and allow people to talk to it via interactive video.

Building a Conversational Friend

I decided to use Cocohub to build a simple conversational friend. I constructed a chatty college student named Megan.

I was impressed by how quick and easy it was for me to create Megan. I’d never used Cocohub previously, but after about 45 minutes I had Megan to the point where I was ready to record this video.

Have a look at the results. Note that I used the phrase “wake up” to kick off the scripted conversation.

My Conversation with Megan

As you can hear from the video interaction, Megan drives the conversation. She’s not a question-answering bot, but more of a chatty personality that asks occasional questions.

The entire conversation is prescripted, with several questions sprinkled in. It’s certainly not state-of-the-art technology for open domain chatbots, but it worked pretty well for my purposes.

No Need to Worry about ASR or NLP

Cocohub did all the heavy-lifting. They handled the voice input audio processing. The Cocohub platform also performed all the NLP tasks. I just needed to select my intents after I opened the mic for the user to talk.

Storing Values

Cocohub offers a lot more capabilities than the ones I used for Megan. For example, it allows you to pick up on certain things a person says and use that information right away, or store it for later use.

For example, I might have had Megan ask “What’s your favorite hobby?” Then I could have easily captured the person’s response and had Megan say “Oh, so you like <hobby>. That’s great.”

Summary

I only scratched the surface of Cocohub’s capabilities with my Megan character, but I found the platform to be very promising. I recommend you try it out for yourself.

Let’s quickly compare the technologies underpinning the conversational avatars I’ve written about so far:

  • Talking Santa by Rapport and Pandorabots — Animated avatar by Rapport, speech-to-text by AWS Transcribe, pre-scripted content by Pandorabots using AIML pattern-matching platform.

  • Nestle Cookie Coach — Animated avatar by Soul Machines, voice assistant NLP programming using Google’s Dialogflow (Intent-based NLU model).

  • Sensorium Virtual Being — Animated avatar by Sensorium, neural network / transformer-based NLG (Natural Language Generation) by Temporal Games.

  • Megan — Animated avatar provided by Cocohub, voice by Amazon Polly, NLP entirely handled by the Cocohub platform, prescripted conversation built by me on Cocohub.

Have fun exploring!

3D, Avatar, Conversational AIFrancesca Tabor