elevenlabs

# Basic TTS — generate audio from text
from elevenlabs import ElevenLabs

client = ElevenLabs(api_key=os.environ["ELEVENLABS_API_KEY"])

# Generate and save audio
audio = client.text_to_speech.convert(
    voice_id="pNInz6obpgDQGcFmaJgB",    # "Rachel" — warm, professional
    text="Welcome to Bright Smile Dental. How can I help you today?",
    model_id="eleven_turbo_v2_5",         # Optimized for low latency (~200ms)
    voice_settings={
        "stability": 0.6,                 # Lower = more expressive, higher = more consistent
        "similarity_boost": 0.8,           # How closely to match the original voice
        "style": 0.3,                      # Style exaggeration (0-1)
        "use_speaker_boost": True,         # Enhance clarity
    },
)

# Save to file
with open("greeting.mp3", "wb") as f:
    for chunk in audio:
        f.write(chunk)

# Streaming TTS — for real-time applications
audio_stream = client.text_to_speech.convert_as_stream(
    voice_id="pNInz6obpgDQGcFmaJgB",
    text="Let me check our available appointments for next Tuesday.",
    model_id="eleven_turbo_v2_5",
    output_format="pcm_24000",            # Raw PCM for WebRTC/LiveKit
)

for chunk in audio_stream:
    send_to_audio_output(chunk)            # Stream directly to speaker

# Instant voice clone — from a single audio sample
voice = client.voices.add(
    name="Dr. Smith",
    files=[open("dr_smith_sample.mp3", "rb")],
    description="Calm, authoritative male voice for medical context",
    labels={"use_case": "voice_agent", "language": "en"},
)
print(f"Cloned voice ID: {voice.voice_id}")

# Professional voice clone (higher quality, requires consent)
# Needs 30+ minutes of clean audio for best results

# ElevenLabs Conversational AI — fully managed voice agent
from elevenlabs import ConversationalAI

agent = ConversationalAI(
    api_key=os.environ["ELEVENLABS_API_KEY"],
    agent_id="your-agent-id",             # Created in ElevenLabs dashboard
)

# WebSocket connection for real-time conversation
async def handle_call(websocket):
    async for audio_chunk in websocket:
        # Send caller audio to ElevenLabs
        response = await agent.process_audio(audio_chunk)
        # Send AI response audio back to caller
        await websocket.send(response.audio)

// Browser-based TTS
import { ElevenLabsClient } from "elevenlabs";

const client = new ElevenLabsClient({ apiKey: process.env.ELEVENLABS_KEY });

// Stream audio in browser
const response = await client.textToSpeech.convertAsStream(voiceId, {
  text: "Hello! How can I assist you?",
  model_id: "eleven_turbo_v2_5",
  output_format: "mp3_44100_128",
});

// Play audio using Web Audio API
const audioContext = new AudioContext();
const reader = response.getReader();
// ... decode and play chunks

pip install elevenlabs                    # Python
npm install elevenlabs                    # Node.js