DeepL, known for text translation, now wants to translate your voice

DeepL, widely recognized for its advanced text translation services, has launched a new voice-to-voice translation suite aimed at enhancing communication in various settings. This innovative suite caters to diverse use cases including business meetings, web and mobile conversations, and group interactions, particularly beneficial for frontline workers via tailored applications. In conjunction with this release, DeepL is introducing an API that empowers external developers and businesses to leverage its technology for customized applications, such as those used in call centers. CEO Jarek Kutylowski emphasized the evolution of the company by stating, "After spending so many years in text translation, voice was a natural step for us." He acknowledged that while DeepL has made significant strides in text and document translation, there was a notable gap in effective real-time voice translation solutions. Creating a product that translates speech in real-time presents unique challenges, particularly in minimizing latency—the delay between the speaker’s words and the translated audio output. DeepL is also rolling out add-ons for popular platforms like Zoom and Microsoft Teams, enabling users to hear real-time translations of conversations occurring in various languages or to view translated text on screen. Currently in early access, DeepL is inviting organizations to sign up for a waitlist to participate in the program. The new technology supports both in-person and remote conversations, allowing users to join group discussions through a simple QR code scan. Importantly, DeepL's voice translation capabilities can adapt to specific vocabularies, including industry jargon and personal names. Kutylowski remarked on the transformative potential of AI in customer service, noting that a translation layer can help companies offer support in languages where skilled personnel are hard to find and costly to hire. DeepL asserts that it maintains control over its entire voice-to-voice translation framework. The current process involves converting speech to text, translating it, and then converting it back into speech. However, the company aims to create a more efficient end-to-end translation model that bypasses the text conversion stage altogether. As DeepL ventures into this competitive space, it faces challenges from several well-funded startups. For instance, Sanas, which secured $65 million last year, uses AI to adjust a speaker’s accent in real-time, primarily targeting call center operations. Other competitors include Dubai-based Camb.AI, which specializes in speech synthesis and translation for media, and Palabra, which aims to develop a real-time speech translation engine that preserves the speaker’s voice and intent, intensifying the competition with DeepL's offering.