News radio is quickly becoming old-fashioned. At a time when video, music, and print are all evolving, audio broadcast news has largely stood still. An innovative new app from a legacy media company is about to change that.
Newsbeat is a new app from Tribune Digital Ventures that uses machine learning and text-to-speech synthesis to remake how people consume news. Using content from the Tribune Company’s hundreds of newspapers, Newsbeat creates an on-demand news radio service tailored to each user.
Roughly 7,000 stories each day will get analyzed for importance and popularity, with the top ones getting read and recorded by humans. The rest of the stories will be put through high-quality text-to-speech translation and be available alongside the others.
Opening the app, you’re greeted by computerized newscasters before going into the top headlines. The app feels like a music playlist that you can skip forward and back as well as zoom out and see all stories and jump right to any specific one.
How does it work? In broad strokes, the stories are scanned continuously throughout the day, categorized by topic and then searched for keywords. They are then de-duplicated and separated from all other similar stories. Next, the stories are selected for users based on the information provided to the app. Things like: location, choice of publication, areas of interest, and the length of one’s commute.
Right now, news radio is still a mostly one-dimensional object. You tune in, listen for however long you have, and then tune out. The listener gets very little say in the editorial process, but Newsbeat’s underlying technology may change that.
Before the app was available to test on my device, president of Tribune Digital Ventures, Shashi Seth played a demo of a few stories over the phone. The difference between a human-read article and a computer-read one was hard to distinguish. It wasn’t perfect, but it wasn’t distracting either.
Text-to-speech has been a critical part of the project since its inception. The team realized it wouldn’t be able to generate enough human power to read all the stories needed, so it partnered with some of what it says are the leading companies in the field. You can make educated guesses as to the companies in question, however, Seth declined to officially say who is involved.
“Right now, things like Siri speak in short one- or two-sentence answers. No one else is really doing anything like this,” says Seth. “We’re aiming for the highest quality reading, something that shows emotion like laughter or inflection, though we’re not quite there yet.”
The text-to-speech feature also gives Newsbeat the ability to provide automated breaking news in audio form. Once a story has been written, it can sent to the app, versus having to have the story read and recorded.
After getting a chance to use Newsbeat for a small amount of time, it’s hard to argue that this type of technology doesn’t at least play some part of the future of news media. The computerized voices can be picked out when specifically listening for them, but otherwise, they blend and the news stories are the biggest focus.
Hearing Newsbeat makes current consumer-facing artificially intelligent voices, even Siri, seem like a joke. This is the new standard for apps that use speech to text.
Swell is another app that’s striving to not make the listener have to choose what to listen to on their commute. Unlike Newsbeat, Swell gathers content already in audio form such as newscasts and podcasts. A listener chooses what topics they’re interested in and Swell does the rest.
You might tell Swell what you’re interested in, but it’s all about the skips and engagement. How long you listen to different segments and which types of stories and publications you skip ultimately begins to weigh heavily on which content you’re hearing.
Newsbeat works the same way.
You’re able to add your home and work addresses, which gives the app an approximation of how many stories you’ll need automatically added to your playlist. So if you have a 10-mile commute that takes 45 minutes because of traffic, the app will account for that as well and provide traffic and weather updates along the way.
Newsbeat uses its own proprietary learning algorithm to pick up on a user’s implicit actions. Even if you don’t add your addresses, it will figure out the length of a person’s daily commute and adjust for those times. If someone doesn’t provide preferred publications, the app will first assume the person wants content from local sources and serve those up along with national publications.
It will also add items that people are most passionate about like information about local sports teams. There will be future capabilities to alert fans when their team goes ahead or the score changes in a close game.
Seth declined to comment on how long the project has been in development, casually mentioning that it hasn’t been long and that he just started at Tribune Digital Ventures about nine months ago. He did say that the team has been working on and tuning the algorithm since the project was started. He emphasized the importance of being able to churn through the data quickly and accurately.
“Are we awesome yet?” says Seth, in reference to the app’s machine learning. “Not yet, but we’re good, and we’re going to be awesome.”
Based on the scale and possible implications of a project like this–including the backing from such a large news organization looking to be well positioned for the future–Newsbeat should be taken very seriously by outlets like NPR.
Like all apps or services that promise incredible experiences, Newsbeat’s success hinges on whether the backend technology is good enough to deliver. If stories aren’t picked well, or more importantly, if the computer-generated speech isn’t acceptable for enough users, then it won’t matter how cool any of the other features are.
Could this change the distribution of news? Quite possibly. At launch, Seth appears to be on to something.