ElevenLabs integration

Overview

ElevenLabs provides AI voice generation – text-to-speech, voice cloning, dubbing, and speech-to-speech. The Automize connector covers TTS (sync + streaming), STT, dubbing jobs, voice management, and history download.

• Regions: Single global API at api.elevenlabs.io. Audio responses stream from the closest edge POP automatically.
• Plans: Free has 10k characters/month and no commercial use rights. Starter onwards unlocks commercial licensing. Voice cloning (Instant + Professional) requires Creator or higher. Dubbing Studio is metered separately.

Setting up the connection

Auth type: API Key.

1. In ElevenLabs, open Profile → API Keys and click Create API Key. Scope it to the workspace + actions you need.
2. Copy the key (shown once) and paste it as the API key in Automize.
3. The connector sends it as xi-api-key on every request – no Authorization header used.

Find your credentials at https://elevenlabs.io/app/settings/api-keys.

Rate limits

Character throughput is the actual quota – concurrent requests cap at 2 (Free), 5 (Starter), 10 (Creator), 15 (Pro+). Hitting concurrency returns 429 system_busy; the connector backs off + retries. Streaming TTS counts the same characters whether you consume the stream or not.

Data model

ElevenLabs' content graph:

    Voice         pre-built / cloned voice
      ├─ settings (stability, similarity_boost, style, speaker_boost)
      └─ samples  (source audio for clones)

    History       every generation gets a history_item_id
      ├─ text     (the prompt)
      ├─ voice_id (which voice rendered it)
      └─ audio    (downloadable for the retention window)

    Dubbing       async job – translate + voice-clone source video
      └─ status   dubbing → dubbed → (failed)

    Model         eleven_multilingual_v2 (default), eleven_turbo_v2_5
                  (low-latency), eleven_monolingual_v1 (legacy English),
                  scribe_v1 (STT). Each has supported language list.

Voice categories:

    premade       ElevenLabs-hosted stock voices, available to all
    cloned        Your Instant Clones (1-min source, fast)
    generated     Voice Lab generated (style-from-description)
    professional  Studio-trained, manual review (Pro+)

History retention: 30 days for Free, longer on paid plans. Audio
is regenerable if you have history retention – past that window it's
gone, even if you have the history_item_id.

Available operations

15 operations available. Click any row to jump to its detail.

Operations

Add Voice (Clone)

Clone a voice from audio samples

Tips

• Instant cloning needs 1+ minute of clean source audio. Professional cloning needs 30+ minutes and a manual review step on ElevenLabs' side.
• Multipart upload: name + description + files[] (audio samples). Returns voice_id; available immediately for Instant clones, hours-days for Professional.

Create Dubbing

Dub a video or audio file to another language

Tips

• Async – returns dubbing_id. Poll Get Dubbing until status='dubbed' (typically 1-5 min/minute of source).
• target_lang takes ISO-639 codes; source_lang is auto-detected if omitted.

Delete Voice

Delete a voice from your library

Tips

• DELETE /v1/voices/{voice_id} – only your cloned voices are deletable. Premade voices return 403.
• Doesn't delete historical generations made with this voice – those stay in history.

Edit Voice Settings

Update default stability/similarity settings for a voice

Tips

• POST /v1/voices/{voice_id}/settings/edit – body: stability, similarity_boost, style, use_speaker_boost.
• Settings persist as the voice's default. Per-call settings (passed in TTS) override these without changing the saved default.

Get Dubbing

Get status of a dubbing job

Tips

• Poll /v1/dubbing/{dubbing_id} every 30s. status: dubbing → dubbed → failed.
• On 'dubbed', call the separate GET /v1/dubbing/{dubbing_id}/audio/{language_code} to download the rendered audio per target language.
• error_message exposed on 'failed' – common: source_audio_quality_too_low, lipsync_failed, unsupported_source_language.

Get History Audio

Download the audio for a historical generation

Tips

• GET /v1/history/{history_item_id}/audio – re-downloads the rendered audio for a past generation, no re-billing.
• Available only if history retention window hasn't expired (30 days Free, indefinite on paid).

Get User

Get account info, subscription tier and character usage

No input parameters.

Tips

• GET /v1/user – character quota remaining, subscription tier, next_character_count_reset_unix, voice_limit. Useful as a health-check + cost-budget guard.

Get Voice

Get a single voice's details

Tips

• GET /v1/voices/{voice_id} – full metadata including category (premade|cloned|generated|professional), samples, settings, sharing options.
• Pass with_settings=true to include the active voice_settings object.

List History

List historical generations

No input parameters.

Tips

• Paginated. Default page_size=100, max 1000. last_history_item_id (cursor) for further pages.
• Filter by voice_id + start_after_history_item_id. source filter (TTS|STS) separates Speech-to-Speech from Text-to-Speech generations.

List Models

List available TTS models

No input parameters.

Tips

• Each model exposes: model_id, name, can_do_text_to_speech, can_do_voice_conversion, languages[], max_characters_request, token_cost_factor.
• Pick the model with the lowest token_cost_factor that supports your language + use case to minimise character billing.

List Voices

List voices available to your account

No input parameters.

Tips

• Returns all voices accessible to the workspace – premade + your clones. Filter client-side by category.
• Use this to populate dropdowns; voice_id is opaque and won't change for premade voices.

Speech to Speech

Convert source audio to target voice

Tips

• POST /v1/speech-to-speech/{voice_id} – upload source audio, get back the same speech in the target voice. Preserves emotion and timing better than transcribe→TTS.

Speech to Text

Transcribe audio to text

Tips

• POST /v1/speech-to-text – supports 99 languages. model_id=scribe_v1 is the production STT model.

Text to Speech

Synthesize speech from text for a given voice

Tips

• POST /v1/text-to-speech/{voice_id} – body is text + voice_settings (stability, similarity_boost, style, use_speaker_boost) + model_id.
• model_id defaults to eleven_multilingual_v2. Use eleven_turbo_v2_5 for ~3x lower latency at slightly lower fidelity.
• output_format=mp3_44100_128 is the default. Use pcm_44100 for downstream processing (no encode pass).

Text to Speech (Stream)

Synthesize speech and stream audio chunks

Tips

• Same endpoint with /stream suffix – returns chunked audio. The connector reassembles to a single file.
• Use optimize_streaming_latency=3 for the lowest time-to-first-byte on Turbo.

FAQ

Why is my voice clone sounding nothing like the source?

Source audio quality dominates. Mono, 44.1kHz, no background music, single speaker, 30-180s of dry speech. Stereo or noisy source produces a voice that sounds like a different person.

Streaming vs sync – when does it matter?

Sync is fine for batches (podcasts, articles). Streaming matters when a human is waiting – voice agents, real-time captioning. Turbo + streaming + optimize_streaming_latency=3 gets you sub-500ms time-to-first-audio.

How do I prevent voices from drifting between calls?

Pin voice_settings.stability=0.5+, similarity_boost=0.75+, and use a deterministic seed (model-dependent). Drift on the same voice usually means voice_settings weren't sent – the API uses your default workspace settings when omitted.

Can I get word-level timestamps back?

Yes – use /v1/text-to-speech/{voice_id}/with-timestamps (separate variant). The connector exposes this; output includes a character→timing map for sync to video.

Related connectors

• Heygen
• Replicate
• Openai