Skip to main content

ElevenLabs

High-quality, low-latency TTS - and Scribe speech-to-text with realtime streaming.

TTS​

import { ElevenLabsTTSProvider } from '@llmrtc/llmrtc-provider-elevenlabs';

const tts = new ElevenLabsTTSProvider({
  apiKey: process.env.ELEVENLABS_API_KEY,
  voiceId: '21m00Tcm4TlvDq8ikWAM',
  modelId: 'eleven_flash_v2_5',
  format: 'mp3'
});

STT (Scribe)​

ElevenLabsScribeProvider covers both transcription modes:

  • Batch - transcribe() posts the utterance to the Scribe v2 API (high accuracy, 90+ languages).
  • Realtime streaming - transcribeStream() connects to the Scribe v2 Realtime WebSocket and yields partial transcripts with sub-150ms latency while the user is still speaking, then a committed final.
import { ElevenLabsScribeProvider } from '@llmrtc/llmrtc-provider-elevenlabs';

const stt = new ElevenLabsScribeProvider({
  apiKey: process.env.ELEVENLABS_API_KEY,
  languageCode: 'en' // optional; auto-detects when omitted
});

Pair it with streamingSTT: true on the server for live interim transcripts - see Streaming Speech-to-Text.

CLI mode:

STT_PROVIDER=elevenlabs-scribe
STREAMING_STT=true                # optional: live interim transcripts
ELEVENLABS_STT_MODEL=scribe_v2    # optional batch-model override

Env vars

  • ELEVENLABS_API_KEY
  • Optional: ELEVENLABS_STT_MODEL

Notes

  • TTS: eleven_flash_v2_5 is optimized for latency; use eleven_multilingual_v2 for quality, or eleven_v3 for the most expressive delivery.
  • Supports streaming TTS; enable streamingTTS: true in the server.