Deepgram expands Flux to 10 languages with mid-call switching for voice agents

Real-time voice artificial intelligence startup Deepgram Inc. today announced the general availability of Flux Multilingual, an extension of its Flux conversational speech recognition model to 10 languages with real-time language detection and the ability to switch languages mid-call.

The extended model is being pitched by the company as the first multilingual conversational speech recognition model, a category it argues is distinct from traditional automatic speech recognition because it’s built for dialogue flow rather than transcription. Flux Multilingual supports English, Spanish, French, German, Hindi, Russian, Portuguese, Japanese, Italian and Dutch in a single model and application programming interface.

The release addresses the issue whereby voice agents that operate across different markets often require developers to stitch together separate transcription models, language detection layers and routing logic, an approach that introduces latency and brittle handoffs between systems. Flux Multilingual replaces that stack with one perception model that can take a language hint from the developer or auto-detect on the fly. It supports native code-switching when a speaker moves between languages inside the same conversation.

The model uses model-based turn detection rather than silence detection to determine when a speaker has finished talking, with end-of-turn decisions delivered in under 400 milliseconds.

Deepgram says the system delivers monolingual-grade accuracy across all 10 supported languages and handles interruptions natively. The release is backward-compatible with existing Flux API integrations and is available through Deepgram’s cloud API or as a self-hosted deployment, with European Union endpoints and software development kits.

“Voice AI agents will soon become the default for how global enterprises interact with customers,” said Deepgram co-founder and Chief Executive Scott Stephenson. “Flux Multilingual gives developers a single perception model to build global voice agents, with the ability to switch languages mid-call. Now, enterprises can deliver the same seamless experience to any customer, in any market.”

According to Deepgram, more than 200,000 developers and 1,300 organizations use its platform, which spans speech-to-text, text-to-speech and full speech-to-speech capabilities. The company has processed more than 50,000 years of audio and transcribed more than 1 trillion words to date.

Flux Multilingual is generally available now, with Deepgram offering a limited-time promotional rate on streaming speech-to-text covering both Flux Multilingual and its Nova-3 model.

Deepgram is a venture capital-backed startup that has raised approximately $216 million in funding, including a round of $130 million in January. Investors in the company include Madrona Venture Group, Tiger Global Management, Wing Venture Capital, Y Combinator Management, Nvidia Corp., Citi Ventures, SAP SE and Goldman Sachs Asset Management.

Image: Deepgram