Paris-based artificial intelligence startup Gladia SAS, developer of AI transcription and audio intelligence services, today announced the launch of Solaria, a state-of-the-art AI model designed for real-time multilingual communications.

Although many businesses outsource transcription and translation for call centers and other business uses to save on cost, it’s becoming increasingly necessary to build real-time support to handle global customer bases. According to a 2023 market report from language industry analyst Slator, 49% of executives surveyed worldwide said they saw financial losses from language barriers.

Gladia said it built Solaria to deliver industry-leading results in speech recognition with high accuracy at ultra-fast speeds compared with the competition in the market. Company benchmarks have shown the AI is capable of an average word accuracy rate of 94% — the highest in the industry — for English, Spanish, French and other common languages.

When a user starts talking, its fastest time to the first word is around 270 milliseconds, making it one of the most responsive speech-to-text models in the industry. This is about the time it takes when speaking to Apple Inc.’s Siri or “Hey, Google,” and how long the user has to wait for the first words to appear on the screen.

This also demonstrates how quickly the AI reacts when it’s interrupted mid-sentence, allowing it to quickly adjust and react. The lower the latency, the more fluid a conversation it can have with the user. Deepgram Inc.’s platform is the only competitor with a shorter latency, at a 223-millisecond response.

The AI delivers complete transcripts in just 698 milliseconds, which is almost half a second faster than most competitors. Deepgram takes an average of 1040 milliseconds, while Speechmatics takes 1158 milliseconds.

“Speech is the most natural way to connect with the world — for the first time, automated speech recognition is closing the divide, enabling humans and AI to truly speak the same language,” said Jean-Louis Quéguiner, Gladia’s co-founder and chief executive.

Gladia said Solaria is built to handle 100 languages, including support for 42 underserved languages not matched by its competitors. The company’s team included native-level accuracy for high-population markets and regional languages common to call-center outsourcing hubs such as Tagalog, Bengali, Punjabi, Tamil, Urdu, Persian and Marathi. It also covers emerging voice markets such as Hatian Creole, Maori, Javanese and Malagasy.

The company built the AI to adapt and learn industry-specific terminology so that it can fit into business-critical operations and understand employee speech patterns including medical or financial jargon. The AI is also able to process speech in loud or noisy environments, such as those that exist in cluttered call centers ensuring high accuracy.

“With Solaria, we have made a breakthrough in AI-powered voice technology that unlocks new opportunities for businesses, driving efficiency and delivering more seamless, impactful customer experiences across diverse languages and markets,” added Quéguiner.

As part of its launch, Gladia announced a strategic partnership with LiveKit, an open-source developer framework for real-time AI voice agents. This will enable developers to use Gladia’s application programming interface to build voice conversational agents with built-in multilingual translation capabilities for AI-powered applications.

Since launching its first transcription and audio intelligence API in 2023, Gladia has gained notable traction in the enterprise market, particularly for meeting recorders and note-taking assistants. The company’s platform is now used by more than 700 customers globally, including Attention Inc., Circleback Inc., Method Financial Inc., Recall AI Inc., Sana Labs AB and VEED.IO Ltd.

Images: Gladia