elevenlabs

ElevenLabs API Skill

Overview

Expert skill for text-to-speech, voice cloning, sound effects, and audio AI using ElevenLabs - the most advanced voice AI platform.

API Key

# API ключи: ~/.claude/.credentials.master.env
# Переменная: ELEVENLABS_API_KEY
ELEVENLABS_API_KEY=os.getenv('ELEVENLABS_API_KEY')

Voices

Use a prebuilt ElevenLabs voice, or set ELEVENLABS_VOICE_ID to your own (cloned) voice ID. List available voices via the /v2/voices endpoint.

When to Use ElevenLabs

Best for:

• Text-to-speech (TTS) generation
• Voice cloning (instant & professional)
• Sound effects generation
• Speech-to-speech voice conversion
• Dubbing & localization
• Real-time voice agents

Advantages:

• Most natural-sounding voices
• 70+ languages support
• Ultra-low latency (75ms with Flash)
• Voice cloning from 1 minute of audio
• Emotional expression control

Dependencies

pip install elevenlabs

Models

Model	ID	Latency	Best For
Eleven v3	eleven_v3	Higher	Highest quality, emotions
Multilingual v2	eleven_multilingual_v2	Medium	29 languages, expressive
Flash v2.5	eleven_flash_v2_5	~75ms	Real-time apps
Turbo v2.5	eleven_turbo_v2_5	~250ms	Balance quality/speed

Basic Usage

Setup Client

from elevenlabs import ElevenLabs
import os

client = ElevenLabs(api_key=os.getenv('ELEVENLABS_API_KEY'))

Text-to-Speech

def text_to_speech(text: str, voice_id: str = "JBFqnCBsd6RMkjVDRZzb",
                   model: str = "eleven_multilingual_v2"):
    """Generate speech from text."""

    audio = client.text_to_speech.convert(
        text=text,
        voice_id=voice_id,
        model_id=model,
        output_format="mp3_44100_128",
        voice_settings={
            "stability": 0.5,
            "similarity_boost": 0.75,
            "style": 0,
            "use_speaker_boost": True
        }
    )

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

    return "output.mp3"

Streaming TTS (Real-time)

def stream_speech(text: str, voice_id: str):
    """Stream speech for real-time playback."""

    audio_stream = client.text_to_speech.stream(
        text=text,
        voice_id=voice_id,
        model_id="eleven_flash_v2_5",  # Best for streaming
        output_format="mp3_44100_128"
    )

    # Play directly
    from elevenlabs.play import play
    play(audio_stream)

Instant Voice Cloning

from io import BytesIO

def clone_voice(name: str, audio_files: list):
    """
    Clone voice from audio samples.

    Requirements:
        - Minimum 1 minute of clean audio
        - Best results with 2-3 minutes
    """
    files = [BytesIO(open(f, "rb").read()) for f in audio_files]

    voice = client.voices.ivc.create(
        name=name,
        files=files,
        remove_background_noise=True
    )

    return voice.voice_id

Sound Effects Generation

def generate_sound_effect(description: str, duration: int = 10):
    """
    Generate sound effect from text description.

    Max duration: 22 seconds
    """
    audio = client.text_to_sound_effects.convert(
        text=description,
        duration_seconds=duration,
        prompt_influence=0.5
    )

    with open("effect.mp3", "wb") as f:
        for chunk in audio:
            f.write(chunk)

    return "effect.mp3"

# Examples:
# "Cinematic braam for movie trailer"
# "Forest ambience with birds chirping"
# "Spaceship engine humming"

Speech-to-Speech (Voice Changer)

def change_voice(audio_path: str, target_voice_id: str):
    """Convert voice in audio to target voice."""

    with open(audio_path, "rb") as audio_file:
        result = client.speech_to_speech.convert(
            voice_id=target_voice_id,
            audio=audio_file,
            model_id="eleven_multilingual_sts_v2",
            remove_background_noise=True
        )

    with open("voice_changed.mp3", "wb") as f:
        for chunk in result:
            f.write(chunk)

    return "voice_changed.mp3"

List Available Voices

def get_voices():
    """Get all available voices."""

    voices = client.voices.search(page_size=100)

    for voice in voices.voices:
        print(f"{voice.name}: {voice.voice_id}")

    return voices.voices

Voice Settings

Parameter	Range	Description
stability	0-1	Voice consistency (0.5 recommended)
similarity_boost	0-1	Voice matching (0.75 recommended)
style	0-1	Style exaggeration (0 recommended)
use_speaker_boost	bool	Enhance voice similarity

Output Formats

• mp3_22050_32 - Low quality
• mp3_44100_128 - Standard (default)
• mp3_44100_192 - High quality (Creator+)
• pcm_16000, pcm_44100 - Raw PCM
• opus_48000_64 - Opus codec

API Pricing

Plan	Cost	Credits
Free	$0/mo	10,000 chars
Starter	$5/mo	30,000 chars
Creator	$22/mo	100,000 chars
Pro	$99/mo	500,000 chars

Quick Reference

Task	Code
TTS	client.text_to_speech.convert(text, voice_id)
Stream	client.text_to_speech.stream(text, voice_id)
Clone voice	client.voices.ivc.create(name, files)
Sound effects	client.text_to_sound_effects.convert(text)
Voice change	client.speech_to_speech.convert(voice_id, audio)

Tips

1. eleven_flash_v2_5 - лучший для real-time (75ms latency)
2. eleven_multilingual_v2 - лучшее качество для записей
3. Для voice cloning нужно минимум 1 минута чистого аудио
4. stability=0.5 - оптимальный баланс эмоций
5. Sound effects до 22 секунд максимум
6. Поддержка 70+ языков