Add voice to your AI app with ElevenLabs

Prerequisites

Step-by-Step

1. 1

   Install the SDK

   The official elevenlabs package handles auth, streaming, and voice management.

   pnpm add elevenlabs

2. 2

   Pick a voice

   Browse the voice library or list your voices via API. Each voice has a stable ID.

   import { ElevenLabsClient } from 'elevenlabs';
   const el = new ElevenLabsClient({ apiKey: process.env.ELEVENLABS_API_KEY });
   const voices = await el.voices.getAll();
   console.log(voices.voices.map((v) => ({ id: v.voice_id, name: v.name })));

3. 3

   Synthesize speech

   Use the streaming endpoint and pipe to a file or directly to a client. eleven_turbo_v2_5 is the cheap fast model.

   import fs from 'node:fs';
   const stream = await el.textToSpeech.convertAsStream('21m00Tcm4TlvDq8ikWAM', {
     text: 'Welcome to the show.', model_id: 'eleven_turbo_v2_5'
   });
   stream.pipe(fs.createWriteStream('out.mp3'));

4. 4

   Stream to a websocket

   For voice agents, use the websocket endpoint. You stream text in as the LLM produces it; audio streams out for sub-300ms latency.

   const ws = el.textToSpeech.convertWithWebsocket(voiceId);
   ws.send({ text: 'Hello', try_trigger_generation: true });

5. 5

   Clone your voice

   Upload 1-3 minutes of clean audio. Clones are ready in seconds.

   const v = await el.voices.add({ name: 'My Voice', files: [fs.createReadStream('sample.mp3')] });
   console.log(v.voice_id);

6. 6

   Tune voice settings

   stability + similarity_boost are the levers. Low stability = expressive but inconsistent. High = monotone but reliable.

Common Pitfalls

What's Next

• ->Pair with the Whisper transcription API for full duplex voice agents.
• ->Use Conversational AI for end-to-end voice apps.

More Add voice to your AI app