The four hosts of Morning Edition sit in the recording studio.
The hosts of ‘Morning Edition’ (Mike Morgan/NPR)

What happens when an NPR host riffs on voice skill scripts

Rebecca Rolfe
Design at NPR
Published in
4 min readSep 4, 2019

--

NPR recently launched the ability to play Morning Edition on Amazon voice-based devices. (Just say, “Alexa, play Morning Edition.”) To build that experience the way the NPR voice team envisioned, a pre-recorded voice had to say some of our planned phrases instead of Amazon’s Alexa.

We thought: Why not ask a host of Morning Edition to do it? Listeners love hearing from Peter Sagal and Bill Kurtis on our Wait Wait Quiz; it makes the experience feel super personal.

Ultimately we needed a Morning Edition host to record two messages: one that wraps up the program, and one that indicates breaking news. In both cases we learned important lessons about the power of the human voice on voice assistants.

Wrap-up message

Talk bubbles showing the path from requesting Morning Edition and the wrap up message playing before the live stream starts.

After a listener hears the end of Morning Edition on their Amazon Echo, the audio switches over to the live stream of that listener’s NPR station. To ease the transition, there’s a message between the end of Morning Edition and the start of the live stream — which could pick up mid-story or even mid-sentence.

The message went through many rounds of rewrites:

  • “That’s all for now. Live from NPR, here’s (station name).”
  • “That was the 7 a.m. hour of Morning Edition. Live from NPR, here’s your local station.”
  • “You just heard a replay of today’s Morning Edition. We’ll now go live to your station.”
The Morning Edition hosts speak into microphones and look at computer monitors.
Steve Inskeep, right, and David Greene in the studio at NPR (Allison Shelley/NPR)

After many conversations with stakeholders, we settled on the last option.

We forwarded the script to the Morning Edition team so they could set up a recording session. David Greene agreed to record for us. (Morning Edition has four hosts, so we’ll be sure to pester the rest eventually.)

We asked David to say, “You just heard a replay of Morning Edition. We’ll now go live to your station.”

He made a straight take:

And he also changed it up a bit. Here’s one where he said: “So you just heard a replay of today’s Morning Edition. Now we’re gonna go live to your station.”

When we got the recordings back, we realized we liked the deviations from the script so much better than the original. Written out, “Now we’re gonna go live,” appeared rather informal for NPR, but it sounded perfectly reasonable when David said it. Had we given the script to another host, they would have devised their own versions in a completely different way that would have also sounded great. It only matters that the words are authentic to the speaker.

Breaking news message

Speech bubbles showing the request to play Morning Edition followed by a possible breaking news message, then the show.

Morning Edition is available on-demand through the afternoon, long after the show airs. But what if news breaks after the show wraps up? It could be a natural disaster or an important announcement from the White House. In that case, a breaking news notice may play before Morning Edition starts.

Here’s David’s recording, where he says, “Just so you know, there is breaking news on NPR right now. To switch to what’s live, say ‘Play NPR.’ ”

Compare David’s version with a robot voice powered by MaryTTS, saying the same thing:

The robot voice is sufficient, but lacks the inflection David brings to it. His tone and pace make you think, “Wow, breaking news! Sounds important!” in a way that pre-patterned voices still struggle with. Platform voices like Alexa, Cortana, Google Assistant, and Siri certainly sound much more realistic than this basic voice, and you can get around some tonal limitations through code. However, it can be a lot of customization for something that comes so naturally from a human.

Compared to robotic voices, the human voice is still incredibly powerful. We’re not sure yet how we’ll balance our amazing journalist talent with default platform voices for our future launches, but this project was the perfect case study for how important the final voice is. Take a listen to our skill, and let us know in the comments how you feel about hearing NPR hosts on your voice-based speakers.

Ironically, when we design for voice platforms, a large part of getting the experience right is more than just phrasing. It’s certainly not about enforcing that phrasing, because the human doing the recording will always infuse their own personality into the message. Giving hosts the freedom to ‘just be yourself’ while on-air has served NPR well, and we should carry that tradition to the NPR voices we bring to voice assistants.

Update: NPR decided to sunset the Morning Edition on-demand initiative on October 1, 2020. When listeners say “Alexa, play Morning Edition,” they will hear a message that it is no longer available and be redirected to the live Member station stream. As no experiment comes with a guarantee of success, we still came away with valuable insights that will help us continue to build and refine our efforts to bring public radio — and our local/national partnership — into the on-demand space.

--

--