Learning out Loud - Alexa Skills Challenge

Client

The Alexa Skills Challenge with Alexa Conversations calls for team to build a skill including the Alexa Conversations dialogue manager, utilising Alexa’s most advanced Alexa voice technologies to help invent the future of Natural Voice Experiences.

Rabbit & Pork teamed up with audio expert Josiah Smithson from The Audio Tailors to create a submission for the competition and hopefully take home the Grand Prize.

Challenge 

With so many children around the world that have been missing out on key time in school, there will be many children having to catch up with their studies and even retake exams over the next year. We wanted to create a skill that would enable them to prepare for exams in a non-screen environment and in a fun way. 

We want to also at the same time bring a rich audio experience to the user, something they might be familiar with when playing other games.

Ideation

We had a number of ideas at the start of the project, including some different to Learning Out Loud. We had a video meeting session where team members gave their ideas. Each idea had to meet the following criteria:

1. could we use Alexa Conversations in the idea?
2.  could this be a voice first experience?
3.  could the skill help in the current pandemic in some way?
4.  would we be able to use rich audio in the experience? 
5.  has this exact idea been done before?

After deliberation, we decided that Learning Out Loud was the best option moving forward. We felt that the customisation nature was something new not seen before.

Solution

How we built it

Testing the conversation tool

With the idea, Jamie tested the new Alexa Conversations to see how it would work alongside our existing knowledge of Alexa Skills. This made sure we were confident the skill would be able to take advantage of the new Alexa Conversations

Happy path scripts

John created a number of happy path scripts, these included a first time user, returning user, correct answers, repeat requests and incorrect answers to see how the conversation flowed. Annotations added marked out dynamic slots, language changes for singular and plurals and most importantly where rich audio could be added.

Briefing The Audio Tailors team for audio

We know from previous experience that skills are much more engaging and memorable with rich audio. Jamie and John sent a brief over to Josiah to create the 13 pieces of audio to be included in the skill.

We decided we needed the following pieces of audio:

• Opening and exiting of the skill
• Calculation sound for after the Alexa conversation
• Intro to the round for different subjects (Maths, English, Science, Geography, History)
• Correct, pass, incorrect and countdown to answer
• End of round

What it does

Learning Out Loud has two elements. First using Alexa Conversations we gain vital information from the player to customise the playing experience. We ask them what year they are in, what subjects they are studying and when their exams will take place.

This then allows us to calculate what type of questions to ask from the database of questions we have.

We then move into the quiz mode where the user is asked 10 questions a round. Each question will be related to the subjects they are studying and adjusted to their year (harder questions for those older).

Each question is multiple choice, the player can answer by either saying A, B, C or D or say the answer just read to them. As the user progresses we keep a track of their score. In future plays of the same subject we’ll give them feedback on what their previous highest score was.

At the end of the round, the user is given their score and then can take another subject or more questions from the same subject.

Development

Development started after the happy path scripts were complete. The skill is built with the Alexa Skills Kit SDK for Node.js, S3 bucket storage for the audio files, Dynamo DB for saving users previous scores and questions

We wanted to be able to add in future questions once deployed and also be able for students to say the answer, e.g. “The san andreas fault” as well as “B”. To do this we used the dynamic entities feature which would allow us to add in new answers slots without the need to update the language model each time.

Testing

We were able to test by sharing access to all members of the team and along with close family and friends with the target audience. We asked people to record the audio for the session and then send it back to allow us to fix any errors. Once amends were created we were able to submit to live.

Icon design

We wanted to give the skill an eye catching icon which would include and educational reference, be easy to recognise and importantly fit well with the format of Alexa skill icons.

We were able to work with Hana Skarratt who created several versions of the skill icon along with a bigger logo.

What’s next for Learning Out Loud

There are a number of areas that we know we can expand on to make this Learning Out Loud even more engaging and increase adoption.

1. Account linking – link up with schools existing software systems so feedback can be given to teachers, which students are using the Skill, questions that students are struggling on, which subjects are students performing well in. we would look to add account linking with popular platforms such as smarttech, Schoology, Edmodo
2. ML / AI generated questions. -educes the need for the teachers to add in questions which is time consuming but feeding a tool with raw text which then creates questions for us
3. Leaderboard systems – allowing users to compete for the top places overall and per subject
4. Reminders API – allowing users to set a reminder to play the game according to when their exam is, e.g. weekly, and then daily nearer to their exam date.