AlexaFeed

TL;DR: Record a phone call & use it to dynamically generate a JSON audio Alexa Flash Briefing feed.

Watch a demo on YouTube.

In more detail

This Flask app creates audio Flash Briefings for Alexa by doing 2 things:

1. Allowing you to call in and record audio clips, saving them to s3 using date-based filename.

2. Generating a dynamic Alexa Flash Briefing JSON feed based on the current date.

Live demo of the feed.

Listen a recording from June 23rd.

Add this skill to your Alexa Flash Briefing and try it live!

Motivation

AlexaFeed was an itch I had to scratch, motivated by a few things:

1. My video project, Wakey Wakey, wasn't getting enough traction and required too much production time for each episode. It was time to explore different formats

2. Blogging is boring, RSS is basically dead, and it seems like nobody reads beyond the headlines anymore.

3. Podcasts are having a moment (again?), but producing & distributing them is this endless chain of steps.

4. Alexa has big market penetration among my friends & family. It's an excellent distribution platform for content, but most people use it for setting timers, listening to music, and checking the weather.

5. Alexa's Flash Briefings are a first class feature, and support both text & audio briefings. And yet, there are very few orgs & content creators targeting this platform.

6. I previously wrote a Jekyll -> Flash Briefing plugin, so you can listen to your blog.

7. Amazon recently introduced metrics for Flash Briefings.

8. I noticed the upcoming Echo Show will support video Flash Briefings...

So, after ruminating on all that, I thought: what if me, Steve, and Richard (my other Wakey Wakey collaborators) could produce new content on our own, and then publish it to Alexa without ever having to touch a computer?

Well, it used to be (and probably still is?) common practice for field reporters to call in their stories to the newsroom.

You've probably you've seen it in The Wire and similar journalism driven movies. Some super excited dude in a bar calls a grizzled newsman in the bullpen and tells him to take all this down. Then, after hanging up, the newsman cranks out a story and submits it before deadline. Neat, right?

So, I created a digital newsroom, where approved reporters can call in stories, set an air date, and schedule them. And then, on the appropriate day, content is published automatically to Alexa, where subscribers can "read" the news by asking for their flash briefing.

So I guess it's like an audio newspaper? And with some code tweaks, you could support multiple news items a day, or replace the audio midway through the day for a "late edition."

Calling in Updates

If you are an authorized caller, the app greets and prompts you to set an air date as a 4 digit DTMF input, (so 1031 would be Halloween).

Next, it'll ask you to record up to 3 minutes of audio and replay it for you.

If you are satisfied with your recording, you press 1 and the app will upload the audio (as an mp3) to s3 using the current day 2017-10-31 as the filename.

Once confirmed, the app will tell you to rock on and hangup. And boom, you've recorded a new episode for your flash briefing!!

FYI: The official Alexa Feed spec specifies a particular loudness for audio content and Twilio's recording quality isn't perfect.

Alexa isn't super forgiving when it comes to audio loudness. It's pretty common for audio briefings to play softly because the engineer, didn't you know, engineer it. So, I went down the rabbit hole and incorporated an FFmpeg pipeline to adjust the loudness do some hi/lo pass filtering. It means you'll need to install ffmpeg as a separate buildpack on Heroku - but it also works SUPER WELL.

Emailing in Updates

I also built an endpoint using Mailgun so that approved users can email in audio files. Why? Because maybe you are some master podcaster and already have a workflow to record & master your audio. Also, Twilio isn't free.

Message format:

1. Subject line should be a date in YYYY-MM-DD format.
2. Message must include a single mp3 attachment and nothing else.
3. Everything else in the email will be ignored.

Approved emailers should receive success/failure confirmation.

Not sure if Alexa supports mixed text/audio briefings - but if it does, this could be a fun way to add text clips to your feed.

Recording Updates From Your Browser

Also added a Ziggeo based route /record for recording new episodes directly from your browser. The front end lets you pick a date, record a video, and then post it to to the app, which does all the usual ffmpeg pipeline stuff and then sends you to /post-record where you can confirm that your episode recorded.

Why would you want this? Because if you have a decent computer mic, or don't have a phone, or don't know how to create good audio, or even just want to use existing video files -- this lets you add new episodes without desktop software. It's experimental, but it works :)

The Feed

When you load the feed url, the server will look for an mp3 file based on the current date: YYYY-MM-DD.mp3.

{
  streamUrl: "https://wakey.io/public/mp3/2017-06-14.mp3",
  mainText: "",
  uid: "82a75e11-8ed2-4035-99ac-aeaa20062158",
  titleText: "Wakey Wakey ~ Wed, June 14",
  updateDate: "2017-06-14T05:56:33.0Z"
}

If it can't find a file for that day, it'll play a canned "we're off the air today" message.

{
  streamUrl: "https://wakey.io/public/mp3/offair.mp3",
  mainText: "",
  uid: "090b085e-f208-4431-9459-0bfd5d323888",
  titleText: "Wakey Wakey is off-air right now, check back again soon!",
  updateDate: "2017-06-15T06:05:02.0Z"
}

Since Alexa is primarily a US thing, the app uses Pacific time to determine what day/time it is.

There are a few additional endpoints too:

• /latest - returns a JSON object containing URL of latest episode, date, and a "nice" date string. And yes, it's CORS enabled:

{
  filename: "http://wwaudio.s3.amazonaws.com/audio/2017-06-26.mp3",
  date: "2017-06-26",
  nice_date: "June 26, 2017"
}

• /episodes- returns a playable list of existing episodes and off-air messages. This route is protected with basic username/password which you'll set as BAUSER & BAPASS environment variables.

• /podcast - generates a mostly iTunes compliant podcast feed (missing duration) - but you'll have to edit all the name/category details yourself.

How to Use

1. Create a public s3 bucket for your audio files with audio and original_audio subfolders. Make sure these are publicly readable.

2. Set environment variables for authorized numbers & emailers, Mailgun API keys, and s3 bucket access (FYI, I'd set DEBUG to False on Heroku and True locally):

S3KI=derppppp
S3SK=derpppppppppp

NS=+1XXXXXXXXX
RM=+1XXXXXXXXX
SL=+1XXXXXXXXX

SESSKEY=rand0mstr1ng

NSE=foo@bar.baz
RME=foo@bar.baz
SLE=foo@bar.baz

MAILGUNDOMAIN=https://api.mailgun.net/v3/foobar
MAILGUNKEY=key-xxxx8765309xxxx

PHONE=+1XXXXXXXXX
EMAIL=foo@bar.baz

BUCKET=foobar
AUDIO=audio/
ORIGINAL=original_audio/
FPATH=https://foobar.s3.amazonaws.com/

PODCASTNAME="YOUR PODCAST NAME"
TZ=America/Los_Angeles

DEBUG=False

BAUSER=username
BAPASS=p@ssw0rd

ZIGKEY=XXXXXXXXXXXX

3. Deploy this hot mess to Heroku and install this buildpack for FFmpeg.

4. Point your Twilio number to https://youapp.com/begin_call. And optionally, set up a Mailgun route and forward incoming requests to https://youapp.com/email.

5. Call yourself and record a test message for today.

6. Go to https://youapp.com/episodes to verify your file is available, set for the right day, and playable.

7. Follow these steps to setup your Flash Briefing on Alexa using your app's root URL.

8. If it works, submit it for certification and tell all your friends to subscribe.

9. Keep publishing new content!!