Class STTv2

Deepgram STTv2 using the Flux model for streaming speech-to-text.

This uses Deepgram's V2 API (/v2/listen) which provides turn-based transcription with support for preemptive generation.

Remarks

Key differences from STT (V1):

• Uses TurnInfo events instead of SpeechStarted/Results
• Supports eagerEotThreshold for preemptive LLM generation
• Sends PREFLIGHT_TRANSCRIPT events when eager end-of-turn is detected

Example

import { STTv2 } from '@livekit/agents-plugin-deepgram';

const stt = new STTv2({
  model: 'flux-general-en',
  eagerEotThreshold: 0.5,  // Enable preemptive generation
});

const stream = stt.stream();
stream.pushFrame(audioFrame);

for await (const event of stream) {
  if (event.type === SpeechEventType.FINAL_TRANSCRIPT) {
    console.log(event.alternatives?.[0]?.text);
  }
}
Copy