The streaming wars aren't just about domestic viewers anymore: Netflix, Disney, HBO Max and the like increasingly compete around the world. Around a third of Disney+ subscribers, for instance, are based in India, and the company is looking to further grow its international audience in Europe and Latin America.
Tel Aviv-based startup Deepdub wants to help streaming services accelerate this kind of international rollout by using artificial intelligence for their localization needs. Deepdub, which came out of stealth on Wednesday, has built technology that can translate a voice track to a different language, all while staying true to the voice of the talent. This makes it possible to have someone like Morgan Freeman narrate a movie in French, Italian or Russian without losing what makes Freeman's voice special and recognizable.
Deepdub was founded in late 2019, and has been in conversations with a number of potential Hollywood clients, according to Deepdub CMO Oz Krakowski. "People's jaws are dropping to the floor," he said, adding that industry insiders especially value the ability to adjust the voice of an AI-generated dubbing track. "We call it voice design," he explained.
Deepdub has published an impressive demo video on YouTube. Krakowski recently gave me some more context on how the company's technology works:
- Deepdub currently can dub videos into six different languages, including French, German and Spanish. The company aims to double that number in the next few months, and ultimately wants to be able to localize videos into 60 languages.
- The technology is designed to fully automate the dubbing process, but Krakowski admitted that higher-end projects still require a bit of manual intervention for things like local idioms and brands. For instance, the convenience store chain 7-Eleven doesn't exist in many countries, so employees may have to help the system find an adequate replacement.
- However, this kind of intervention only has to happen once, and it benefits all future projects. "The system keeps learning," Krakowski said. "The amount of manual work decreases over time."
- Deepdub can work with a variety of source materials. This includes complete video files with mixed-down audio tracks from which the system is able to remove and replace individual voices. That could especially come in handy for streaming services with a lot of library content. "You have streaming services with thousands of videos that are stuck on a single language," Krakowski said.
Deepdub is not the only company looking to use AI for dubbing and voice localization. Toronto-based Resemble has built tech to clone your voice in different languages, which is already being used for chatbots and automated call centers. London-based Synthesia also targets Hollywood, albeit with a bit of a different approach: It changes not only the audio track, but actually uses deepfake image manipulation to match an actor's mouth movement to the new voice. Krakowski pointed out that such an approach is still challenged by different camera angles as well as onscreen motion, but he acknowledged that the two technologies may ultimately converge. "It's not a question of if, just of when," he said.
In the meantime, AI dubbing may be able to help Hollywood with some more pressing problems. The pandemic forced streaming services to limit dubbing for a number of markets. For instance, Netflix has been telling viewers in recent months that it had to delay localization for some of its titles. "We're prioritizing voice actors' and dubbing studio safety," the company informed viewers within its app. Artificial intelligence could help to bridge those gaps, even if studios ultimately decide to rely on real voice actors for the final version.
A version of this article will appear in this week's Next Up newsletter. Sign up here to receive it in your inbox every Thursday.