Artificial intelligence (AI) translation corporations DeepL said on the 18th that it will acquire Mixhalo, a real-time, ultra-low-latency audio platform. With the acquisition, the company aims to strengthen real-time voice translation for complex, large-scale events such as international conferences, sports games and concerts.
DeepL will integrate Mixhalo's ultra-low-latency audio infrastructure into its real-time AI voice translation solution, DeepL Voice. The company said, "This allows us to deliver real-time translated audio and captions quickly and clearly, while keeping a natural speaking flow, even at large-scale events with tens of thousands of participants."
Mixhalo was co-founded in 2016 by Mike Einziger, a guitarist and composer for the rock band Incubus, violinist Ann Marie Simpson-Einziger, and CEO and engineer Big Singh. It was launched with the goal of improving concertgoers' listening experience, but later expanded its business into real-time audio services for sports games and various live events.
Based on ultra-low-latency technology, the company simultaneously provides synchronized, high-quality audio to thousands of listeners. To date, it has raised a total of $39 million (about 56 billion won) from major venture capital (VC) firms.
Big Singh, Mixhalo's CEO, said, "Mixhalo delivers the highest-quality audio to every live audience member, regardless of event size, with under 20 milliseconds (ms) of latency," and added, "As we expand into real-time translation, the biggest challenge is finding technology that can keep up with live speech speed while maintaining quality."
Jarek Kutylowski, DeepL's CEO, said, "DeepL Voice is already transforming how humans and corporations communicate across diverse language environments, and together with Mixhalo we can expand this into larger and more complex settings," adding, "The two companies are currently collaborating with the goal of building real-time language AI communication infrastructure."
According to DeepL, DeepL Voice scored 96.4 in quality in an independent benchmark study conducted by Slator this year. In accuracy, naturalness and stability, it outperformed Microsoft Teams, Zoom and Google Meet. In particular, its error rate was 4%, significantly lower than the market average of 17%.