Oslo’s Aftenposten has introduced a cloned voice generated using machine learning to enable users to listen to content.

Product designer Kaya Maraz and senior product manager Lena Beate Hamborg Pedersen of the Norwegian Schibsted masthead tell the story of how this was achieved in an INMA ideas blog.

The AI-assisted system generates an audio file automatically using the cloned voice, when an article is published. If a journalist changes the text, the audio file is automatically updated.

“Since many news stories are updated frequently, we could not provide such a listening service if journalists had to read and record the articles themselves,” says Maraz. “Believing it should be just as easy to listen to news from Aftenposten as it is to read it, we determined offering an audio alternative with our cloned voice is the fastest and least expensive way to do it.”

While there had been a steady increase in listeners, this had not been a high proportion compared to reading numbers. “Now around ten per cent of our subscribers listen at least once a month.”

However, the ambition for the audio alternative is not to see the highest possible numbers, but for the audio articles to supplement reading Aftenposten, rather than substitute for them.

Talking to users, the Aftenposten team learned that readers were choosing to listen in situations where they can’t read. “We have heard several times that people like to listen when folding clothes, exercising, driving, or cleaning the house,” says Maraz. “In other words, they find listening to audio articles useful in situations where they are busy with their hands but still want to be entertained or informed.

They also learned that subscribers really appreciate the opportunity to use their subscription in these situations where reading is not possible. Possibly, by giving subscribers more value for their money, they will subscribe for longer. 

However, they want to use the audio articles in situations where a single audio article is too short to fill the time they have available, such as when driving to work. With most audio articles are between three and four minutes long, it is impossible to press play on each new audio article. So far, they have experimented with “end-to-end” related audio articles, as well as offering curated playlists – topic-based evergreens and a playlist about current affairs.

Aftenposten set a listening record after the launch of these features.

“Compared to the readership, listening figures are not high, but after launching playlists, we broke all previous records,” she says.

After introducing playlists, Aftenposten saw an uptick in daily total listens.

The number of unique daily users has also seen a sharp increase after we launched playlists and automatically play the next audio article

The next graph shows the number of audio articles subscribers who listen on average on a given day. “We saw a clear increase after launching playlists that automatically play next.”

The number of audio articles subscribers who listen on average on a given day increased after the launch of playlists that automatically play next.

The graph below shows the percentage of users who have listened to at least one audio article in the last 30 days. The goal was to surpass having ten per cent of subscribers listen at least once each month, and this has already been reached.

In addition to gathering data, people who tested the playlists were interviewed. Their feedback helped inform what we will focus on for audio in 2024. For more on that, read here.