canonical: https://jentic.com/apis/elevenlabs.io/elevenlabs

# ElevenLabs API

Convert text to natural-sounding speech, clone voices, and process audio across 151 endpoints covering text-to-speech, speech-to-speech, voice generation, sound effects, dubbing, and audio isolation. Supports 29+ languages, custom voice creation from audio samples, streaming audio output, and pronunciation dictionaries. Authentication uses the xi-api-key header for all requests.

## For AI agents

Synthesize speech from text with custom or cloned voices, convert speech between voices, and generate sound effects. Supports 29+ languages with streaming output for real-time audio.

## Scope

Does not handle speech-to-text transcription, natural language understanding, or text generation - use for voice synthesis, audio processing, and sound generation only.

## Capabilities

- Synthesize natural speech from text with 29+ language support and emotion control
- Clone custom voices from uploaded audio samples for personalized synthesis
- Convert speech from one voice to another while preserving content and timing
- Generate sound effects and ambient audio from text descriptions
- Stream audio output in real-time for low-latency playback applications
- Isolate vocals from background noise in audio recordings
- Dub video content into multiple languages with voice matching

## Use cases

### AI Agent Voice Synthesis via Jentic

AI agents discover and invoke ElevenLabs text-to-speech through Jentic's intent-based search to add voice output to conversational interfaces. Agents search for the synthesis operation, receive the input schema including voice_id and model_id parameters, and execute requests that return audio streams. No SDK setup required - agents specify text, voice, and output format and receive MP3 or PCM audio directly.

Example prompt: Search Jentic for 'convert text to speech', load the POST /v1/text-to-speech/{voice_id} schema, and execute with a voice ID and 200-word text input to receive MP3 audio

### Custom Voice Cloning

Create custom voices from audio samples using voice generation endpoints. Upload reference audio files via voice samples, configure voice characteristics, and use the resulting voice_id for all subsequent text-to-speech calls. Supports both instant voice cloning from a single sample and professional voice cloning with higher fidelity from multiple samples.

Example prompt: Upload an audio sample to create a voice clone via POST /v1/voice-generation/create-voice, then synthesize text using the returned voice_id via POST /v1/text-to-speech/{voice_id}

### Real-Time Streaming Speech

Stream synthesized audio in real-time using the POST /v1/text-to-speech/{voice_id}/stream endpoint for low-latency applications. Returns chunked audio as it is generated, enabling playback to begin before the full synthesis completes. Ideal for voice assistants, live narration, and interactive dialogue systems where response time matters more than batch processing efficiency.

Example prompt: Stream a 500-word article as speech via POST /v1/text-to-speech/{voice_id}/stream with output_format 'mp3_44100_128' and begin playback from the first audio chunk

### Audio Isolation and Enhancement

Separate vocals from background noise in audio recordings using POST /v1/audio-isolation. Upload mixed audio and receive a clean vocal track with background sounds removed. Useful for podcast editing, meeting transcription preprocessing, and cleaning noisy recordings before speech-to-text or voice conversion processing.

Example prompt: Submit a noisy audio recording to POST /v1/audio-isolation and retrieve the cleaned vocal track for downstream transcription

### Sound Effect Generation

Generate custom sound effects from text descriptions using POST /v1/sound-generation. Describe the desired sound in natural language and receive synthesized audio matching the description. Supports ambient sounds, foley effects, and musical elements for game development, video production, and interactive media without requiring a sound library.

Example prompt: Generate a 'thunderstorm with heavy rain' sound effect via POST /v1/sound-generation with duration of 10 seconds

## Key endpoints

| Method | Path | Description |
| --- | --- | --- |
| POST | /v1/text-to-speech/{voice_id} | Convert text to speech with a specific voice |
| POST | /v1/text-to-speech/{voice_id}/stream | Stream synthesized speech in real-time |
| POST | /v1/speech-to-speech/{voice_id} | Convert speech from one voice to another |
| POST | /v1/sound-generation | Generate sound effects from text descriptions |
| POST | /v1/audio-isolation | Isolate vocals from background audio |
| GET | /v1/voices | List all available voices |
| POST | /v1/voice-generation/create-voice | Create a voice from generated parameters |
| GET | /v1/history | Retrieve speech generation history |

## Key resources

- **Text-to-Speech** — Synthesize speech from text with voice selection, streaming, and timestamp support
- **Speech-to-Speech** — Convert audio from one voice to another while preserving content
- **Voices** — List, configure, and create voices including clones from audio samples
- **Sound Generation** — Create sound effects and ambient audio from text descriptions
- **Audio Isolation** — Separate vocals from background noise in audio recordings
- **History** — Access past generation outputs and download previous audio files
- **Projects** — Organize long-form audio content into chapters and manage narration workflows

## Why Jentic

- **Setup:** Wiring ElevenLabs by hand means handling its API-key auth, setting the xi-api-key header on every call, and managing voice ids, model choices, and output formats yourself. Through Jentic you install once, import the ElevenLabs API from the API Directory, store the API key once, and your agent calls it.
- **Permission scoping:** ElevenLabs puts the voice id in the URL path (/v1/text-to-speech/{voice_id}, /v1/speech-to-speech/{voice_id}), so a rule can pin your agent to one voice for synthesis. You choose the operations it may call, so voice creation or sound generation are not included unless you add them.
- **Credential handling:** Your ElevenLabs API key is stored once, encrypted, by your own Jentic One instance and injected at execution time. It never enters the agent's prompt, logs, or context.
- **Discovery method:** Agents search Jentic by intent such as 'convert text to speech with a custom voice' or 'isolate audio from a clip', and Jentic returns the matching ElevenLabs operation with its input schema so the agent calls the right endpoint without browsing the reference docs.

## Related APIs

- **OpenAI API** — TTS and Whisper transcription as part of a broader AI model ecosystem
- **AssemblyAI API** — Speech-to-text transcription and audio intelligence to complement voice synthesis
- **Deepgram API** — Real-time speech-to-text with streaming transcription for conversational AI
- **Stability AI API** — Image and video generation to pair with audio for multimedia content creation

## FAQ

### What authentication does the ElevenLabs API use?

The ElevenLabs API uses an API key passed in the xi-api-key header with every request. You can find your key in the Profile tab at elevenlabs.io. Through Jentic, your xi-api-key is stored encrypted in your Jentic One instance and agents receive scoped access without the raw key entering their context.

### Can I clone a custom voice with the ElevenLabs API?

Yes. Upload audio samples via the voice creation endpoints to create a voice clone. Instant voice cloning works from a single short sample via POST /v1/voice-generation/create-voice. Professional cloning requires multiple samples for higher fidelity. The resulting voice_id can be used in all text-to-speech and streaming endpoints.

### What are the rate limits for the ElevenLabs API?

Rate limits are tied to your subscription tier. Free accounts get 10,000 characters per month. Starter plans provide 30,000 characters, and higher tiers scale up to millions. Concurrent request limits vary by plan - the API returns 429 status codes when exceeded. Character usage can be checked via GET /v1/user/subscription.

### How do I synthesize speech through Jentic?

Search Jentic for 'convert text to speech with a custom voice' to discover the POST /v1/text-to-speech/{voice_id} operation. The schema shows required parameters: voice_id in the path and text in the body. Optionally set model_id for multilingual synthesis. Execute through Jentic's SDK (pip install jentic) and receive the audio file directly.

### What audio formats does the ElevenLabs API support?

The text-to-speech endpoints support MP3 (various bitrates: 44100_128, 44100_64), PCM (16000, 22050, 24000, 44100), and mu-law formats. Specify the desired format via the output_format query parameter on POST /v1/text-to-speech/{voice_id}. Streaming endpoints support the same formats for chunked delivery.

### Can I generate sound effects with the ElevenLabs API?

Yes. The POST /v1/sound-generation endpoint accepts a text description of the desired sound and generates matching audio. Describe sounds in natural language like 'gentle rain on a tin roof' or 'car engine starting'. The endpoint returns the generated audio file directly. Useful for game audio, video production, and interactive media.

### What languages does ElevenLabs text-to-speech support?

ElevenLabs supports 29+ languages including English, Spanish, French, German, Italian, Portuguese, Polish, Hindi, Arabic, Japanese, Korean, and Chinese. Use the multilingual_v2 model for non-English synthesis. Language is auto-detected from input text, or you can specify it explicitly in the request body for more accurate pronunciation.

### Can I limit what my agent is allowed to do with the ElevenLabs API?

Yes. Because you self-host Jentic One, your own rules decide which ElevenLabs operations and credentials the agent may use, so you can allow text-to-speech (POST /v1/text-to-speech/{voice_id}) and its streaming variant while excluding voice creation and sound generation until you explicitly add them. Since the voice id sits in the URL path for the synthesis and speech-to-speech endpoints, a rule can pin your agent to a single approved voice. Your xi-api-key is stored once by your own instance and injected at execution time, so the agent never receives the raw key.