Setting Up ElevenLabs
Like a Professional

1. Go to elevenlabs.io and click Sign up (top right).
2. Use a real work email — you will attach billing to it later.
3. Verify the email link, choose a strong password, enable 2FA in Profile → Security.
4. Pick the Free plan to start. You get 10,000 characters/month — enough for the whole chapter.

The left rail is the whole product in miniature. Top to bottom: Home, Voices,Speech, Dubbing, Projects, Sound Effects, Conversational AI, and Usage. For voice automation work you will live in Voices,Speech and Conversational AI.

Projects is ElevenLabs' long-form editor: paste a book chapter, assign voices per paragraph, regenerate line-by-line, and export a stitched WAV/MP3. Perfect for audiobooks and podcasts. For real-time voice agents we won't use Projects — we'll call the API directly. Know it exists.

Two libraries: Default (11 pro voices shipped by ElevenLabs — Rachel, Adam, Antoni, Bella, etc.) and Community (thousands of voices contributed by users). Each voice has a Voice ID — a 20-character string you will paste into your code.

• Instant Voice Clone (IVC) — 1 minute of audio, ready in 30 seconds, 90% quality.
• Professional Voice Clone (PVC) — 30+ minutes of studio audio, trained overnight, 99% quality.

6. Instant Voice Clone (IVC)

1. Record 1–3 minutes of clean audio in WAV or MP3 (16kHz+ mono).
2. Voices → Add a new voice → Instant Voice Clone.
3. Upload up to 25 files, name the voice, tick the consent box.
4. Click Create voice. Wait ~30 seconds. Play the preview.

7. Professional Voice Clone (PVC)

PVC is the model you use when the voice IS the product — an audiobook narrator, a branded IVR persona. Requirements: 30 minutes to 3 hours of clean studio audio, one speaker, one mic, one room. Training takes 4–8 hours. Only available on paid plans.

1. Profile (top-right avatar) → API Keys → Create API Key.
2. Name it after the project: helpdesk-prod, helpdesk-dev.
3. Copy once — you cannot view it again. Paste into a .env file, never into code.
4. Add .env to .gitignore. If you ever push a key: rotate immediately.

# .env
ELEVENLABS_API_KEY=sk_1234...abcd
ELEVENLABS_VOICE_ID=EXAVITQu4vr4xnSDxMaL

• Free — 10k chars/month, 3 custom voices, non-commercial.
• Starter $5/mo — 30k chars, IVC, commercial license.
• Creator $22/mo — 100k chars, PVC, 192 kbps audio.
• Pro $99/mo — 500k chars, 44.1 kHz WAV, usage analytics.
• Scale / Business — custom, priority latency, dedicated support.

1. Speech → Text to Speech.
2. Voice: Rachel. Model: eleven_multilingual_v2.
3. Paste: "Hello. My name is Nova. Welcome to voice automation."
4. Click Generate. Preview. Download the MP3.

• Stability (0–1) — low = expressive/variable, high = monotone/consistent. Start at 0.5.
• Similarity Boost (0–1) — how tightly to match the original voice. Start at 0.75.
• Style Exaggeration (0–1) — pushes the voice's characteristic style. Start at 0.0; raise for characters.
• Speaker Boost — a toggle that sharpens speaker identity. Leave on.

Every voice has a stable ID. Memorize a few, copy the rest from the library.

export const VOICES = {
  Rachel:  "21m00Tcm4TlvDq8ikWAM",
  Adam:    "pNInz6obpgDQGcFmaJgB",
  Bella:   "EXAVITQu4vr4xnSDxMaL",
  Antoni:  "ErXwobaYiN019PkySvjV",
  Domi:    "AZnzlk1XvdvUeBnXmlld",
} as const;

The one endpoint you will hit most is:

POST https://api.elevenlabs.io/v1/text-to-speech/{voice_id}
Headers:
  xi-api-key: <YOUR_KEY>
  Content-Type: application/json
Body:
  {
    "text": "Hello world",
    "model_id": "eleven_multilingual_v2",
    "voice_settings": { "stability": 0.5, "similarity_boost": 0.75 }
  }
Response: audio/mpeg binary stream

1. New request → POST → paste the URL, replace {voice_id} with Rachel's ID.
2. Headers: add xi-api-key = your key.
3. Body → raw → JSON: paste the JSON above.
4. Click Send and Download → save as hello.mp3. Play it.

# pip install requests python-dotenv
import os, requests
from dotenv import load_dotenv
load_dotenv()

KEY = os.environ["ELEVENLABS_API_KEY"]
VOICE = "21m00Tcm4TlvDq8ikWAM"   # Rachel

r = requests.post(
    f"https://api.elevenlabs.io/v1/text-to-speech/{VOICE}",
    headers={"xi-api-key": KEY, "Content-Type": "application/json"},
    json={
        "text": "Hello from Python.",
        "model_id": "eleven_multilingual_v2",
        "voice_settings": {"stability": 0.5, "similarity_boost": 0.75},
    },
    timeout=30,
)
r.raise_for_status()
open("hello.mp3", "wb").write(r.content)
print("wrote", len(r.content), "bytes")

// npm i node-fetch dotenv
import "dotenv/config";
import { writeFileSync } from "node:fs";

const KEY = process.env.ELEVENLABS_API_KEY!;
const VOICE = "21m00Tcm4TlvDq8ikWAM";

const res = await fetch(`https://api.elevenlabs.io/v1/text-to-speech/${VOICE}`, {
  method: "POST",
  headers: { "xi-api-key": KEY, "Content-Type": "application/json" },
  body: JSON.stringify({
    text: "Hello from JavaScript.",
    model_id: "eleven_multilingual_v2",
    voice_settings: { stability: 0.5, similarity_boost: 0.75 },
  }),
});
if (!res.ok) throw new Error(await res.text());
writeFileSync("hello.mp3", Buffer.from(await res.arrayBuffer()));
console.log("done");

curl -X POST "https://api.elevenlabs.io/v1/text-to-speech/21m00Tcm4TlvDq8ikWAM" \
  -H "xi-api-key: $ELEVENLABS_API_KEY" \
  -H "Content-Type: application/json" \
  --data '{"text":"Hello from cURL","model_id":"eleven_multilingual_v2"}' \
  --output hello.mp3

• mp3_44100_128 — default, 44.1 kHz, 128 kbps — good for most cases.
• mp3_22050_32 — smaller, telephony grade.
• pcm_16000 — raw PCM, feed into Twilio / DSP.
• ulaw_8000 — μ-law 8 kHz, real phone lines.

Set with the output_format query param: ?output_format=pcm_16000.

// Copy-paste presets
export const PRESETS = {
  narration:    { stability: 0.55, similarity_boost: 0.80, style: 0.0 },
  conversation: { stability: 0.35, similarity_boost: 0.80, style: 0.15 },
  ivr:          { stability: 0.75, similarity_boost: 0.85, style: 0.0 },
  character:    { stability: 0.25, similarity_boost: 0.75, style: 0.6 },
};

• 401 Unauthorized → wrong or rotated API key. Regenerate; update .env.
• 422 Unprocessable → text too long (max 5000 chars per request) or invalid voice_id.
• 429 Rate Limit → your plan's concurrency exceeded. Back off 1s and retry.
• Robotic output → stability too high. Drop to 0.35.
• Muffled clone → training audio had noise. Re-upload denoised WAV.