legoktm/speakca

Alexa skill for California Speaks (<https://speakca.net/>)

speakca is a custom Alexa skill for California Speaks.

California Speaks is a radio program that invites listeners to hear a question each week, call in their response, and then listen to selected responses the next week.

There are three main features to the skill: hearing the question, listening to previous answers, and searching through previous responses.

User experience

Question

We use regexes to get the question out of the HTML of the California Speaks website. While this might appear hacky, it has been rather reliable so far, and should be reasonable to maintain.

A future optimization might be to add caching here, since the question is only going to change once a week.

Answers

The fetch.py script downloads episodes out of soundcloud and into an S3 bucket. We use the AudioPlayer interface to turn the S3 files into a playlist for Alexa to go through.

Search

We depend upon WordPress's search functionality to search through the recorded transcripts. Then we are able to match each WordPress post back to the audio file that we put into S3.

Backend

The custom skill is powered by the fantastic Flask-Ask library, and is designed to run as an AWS Lamba function. The fetch.py script currently requires ffmpeg to be installed, so it's not easily run in AWS Lambda yet. You'll also need to set it to run on a cron.

The bundle shell script will create a zip file that can be uploaded to the AWS Lambda web interface with all dependencies included for deployment. It currently is around 7MB. It should always stay under 10MB.

License

(C) 2018 Kunal Mehta <[email protected]> under the terms of the AGPL, v3 or any later version. See COPYING for more details.