Language Support Expansion

Real-time transcription in RealtimeKit now supports 10 languages with regional variants, powered by Deepgram Nova-3 running on Workers AI. This enhancement allows developers to build multilingual voice applications without relying on external transcription services.

Architecture and Performance

Participant audio is routed through AI Gateway directly to Nova-3 on Workers AI, keeping transcription on Cloudflare's network end-to-end. This approach significantly reduces latency compared to routing requests through external speech-to-text services, making it ideal for real-time applications.

Supported Languages

The feature supports the following languages with regional variants:

• English: en-AU, en-GB, en-IN, en-NZ
• Spanish: es-419
• French: fr-CA
• German: de-CH
• Portuguese: pt-BR, pt-PT
• Hindi, Russian, Japanese, Italian, Dutch: Standard variants

Additionally, developers can use multi for automatic multilingual detection when the speaker's language is unknown.

Implementation

Setting up language-specific transcription is straightforward. When creating a meeting, specify the language via the ai_config.transcription.language parameter:

{
  "ai_config": {
    "transcription": {
      "language": "fr"
    }
  }
}

Use Cases

This capability is particularly valuable for voice agents and real-time translation workflows, enabling applications to transcribe in the caller's language natively without additional complexity or external integrations.