Skip to content
ESLog inStart free
Voice API · Voyager 1

The voice API you already know

800+ voices in 55+ languages, at 48 kHz, with emotion written right into the text. Works with the ElevenLabs SDK: change the base URL and the key, and your code keeps running.

API on every plan1 credit = 1 characterC2PA on every file
ElevenLabsVoicefinder
from elevenlabs.client import ElevenLabs
client = ElevenLabs(
Removed line: api_key="sk_…",
Added line: api_key="vf_…",
Added line: base_url="https://api.voicefinder.io",
)
audio = client.text_to_speech.convert(
voice_id="f6abdab0-2524-43d1-92e2-b39486f39fbe", # Agatha
text="[upbeat] Hi! You're now talking to Voicefinder.",
model_id="eleven_multilingual_v2",
language_code="en",
)
with open("hello.mp3", "wb") as f:
for chunk in audio:
f.write(chunk)
Two lines change: the key and the base URL. model_id can stay; the voice_id comes from our catalog.
200 OKaudio/mpeg48 kHzx-c2pa-signed: true
Migrate from ElevenLabs

Change the URL. Keep everything else.

The API accepts the field names, headers and formats of the ElevenLabs API. The official SDK, in Python and JavaScript, works without touching anything else.

  1. Base URL

    https://api.voicefinder.io

    No /v1: the SDK adds it.

  2. Key

    vf_…

    In xi-api-key, x-api-key or Authorization: Bearer.

  3. Voices

    client.voices.search()

    Use voice_id values from our catalog: GET /v2/voices.

Equivalents

xi-api-keyxi-api-key · x-api-key · Bearer
text_to_speech.convertPOST /v1/text-to-speech/{voice_id}
voices.searchGET /v2/voices
model_id: eleven_*voyager-1, no error
voice_settings.speed0.85 to 1.25
text_to_sound_effects[doorbell] inside the text
output_format: mp3_* · wav_*MP3 · WAV at 48 kHz

The official SDK works as is

The same package you already have installed. Only the way you create the client changes.

pip install elevenlabsnpm install @elevenlabs/elevenlabs-js

What is different, plainly

  • Every file ships marked, so pcm_*, ulaw_8000 and opus_* return 422: those formats cannot carry the C2PA credential.
  • On Voyager 1, /stream exists so the SDK works, but it does not bring the first byte forward: the file is signed whole before it is sent. For real time, there is Pulsar 1.
Voicefinder vs ElevenLabs, point by point 
Models

Two models. The same voices.

Voyager 1 to produce audio at the highest quality. Pulsar 1 to hold real-time conversations. Same voices, same languages and same emotions in both.

Compare models 
model_id: "voyager-1"

Voyager 1

Our highest-quality model, for when the audio is the product.

  • REST: one call, and the audio comes back in the response
  • MP3 or WAV at 48 kHz
  • Up to 10,000 characters per request
  • Watermark and C2PA on every file
A Voyager 1 call: POST /v1/text-to-speech/{voice_id}
{ "text": "[upbeat] It is here!", "model_id": "voyager-1" }
→ 200 · audio/mpeg · 48 kHz · x-c2pa-signed: true

For narration, advertising, e-learning, dubbing and audiobooks.

Meet Voyager 1 
model_id: "pulsar"Coming soon

Pulsar 1

~100 msto first audio

Server-side median, from 1 to 48 simultaneous conversations. Provisional model_id.

  • WebSocket streaming: speak, flush and cancel
  • Interruptions by context_id
  • PCM, μ-law and A-law, ready for telephony
  • The same voices and all 55+ languages
Pulsar 1 WebSocket messages: { "type": "speak", "text": "Hi, how can I help?", "context_id": "t1" }
{ "type": "flush", "context_id": "t1" }
{ "type": "cancel", "context_id": "t1" }

For voice agents, assistants, telephony and real-time apps.

Endpoints

The endpoints you already know.

Same shape, same paths. The ones marked SDK answer the way the ElevenLabs SDK expects; REST ones have their own documented shape.

Full reference 
MethodPathWhat it doesCompatibility
POST/v1/text-to-speech/{voice_id}Generates speech. The response body is the audio.SDK
POST/v1/text-to-speech/{voice_id}/streamThe same audio, delivered in chunks.SDK
GET/v1/voicesThe whole catalog, in one page.SDK
GET/v2/voicesThe catalog with filters and token pagination.SDK
GET/v1/voices/{voice_id}One voice with its samples, styles and languages.SDK
POST/v1/voices/addClone your voice from a recording. Private, with a signed rights statement.REST
GET/v1/modelsVoyager 1, its languages and the per-request cap.SDK
GET/v1/historyYour generations. Download them again at no cost.REST
GET/v1/user/subscriptionYour plan and your characters.SDK
GET/v1/sound-effectsThe sound effects you can write into the text.REST

SDK = compatible with the official ElevenLabs SDK · REST = our own shape, in the docs. Base: https://api.voicefinder.io

In every response

What every call brings back.

No extra parameters, no higher plan: all of this comes with every generation, on the free plan too.

POST /v1/text-to-speech/f6abdab0-2524-43d1-92e2-b39486f39fbe200 OK
// example request
{ "text": "[doorbell] [upbeat] Coming! Your order just arrived.",
"model_id": "eleven_multilingual_v2", "language_code": "en" }
content-type:audio/mpeg
x-output-format:mp3_48000_192
x-model-id:voyager-1
x-model-id-warning:model_id "eleven_multilingual_v2" mapped to voyager-1
x-language:en
x-sfx-applied:doorbell:3
x-watermarked:true
x-c2pa-signed:true
request-id:e7c1a9b2-5d4f-4a38-9c0e-6f2b8d7a1c53
x-character-count:32
character-cost:32
Real API headers; example values.Verify an audio file →
MCP connector

Your agent can talk now.

Connect Claude, Cursor or any MCP client to your account: search voices, write the script into a studio project, get a quote and generate. With your key, your credits and the same marking as the studio.

  • studio_voices
  • studio_apply_script
  • studio_estimate
  • studio_generate
  • studio_takes
  • studio_credits
  • Signed quoteGenerating requires the quote_id from studio_estimate and confirm: true. Change one comma in the script and the quote is void.
  • No double chargesRetrying with the same quote_id returns what was already generated instead of generating it again.
  • API limits applyEvery generation takes a slot on your account and every call shows up in your usage.
claude mcp add --transport http voicefinder \
https://api.voicefinder.io/mcp \
--header "Authorization: Bearer YOUR_API_KEY"
For claude.ai connectors, which only ask for a URL: https://api.voicefinder.io/mcp/YOUR_API_KEY. That URL contains your key: treat it like a password.
Limits and pricing

API on every plan.

1 credit = 1 character, from the same balance as the studio. About 972 characters make a minute of audio. Pay yearly and save 30%.

See all pricing 
Free€010,000 credits a month · personal use
Simultaneous generations
1
Generations per minute, per key
10
Characters per hour
20,000
Basic · Plus · Profrom €11/mo100,000 to 1.2M credits a month
Simultaneous generations
3
Generations per minute, per key
30
Characters per hour
200,000
Growth · Business · Enterprisefrom €269/moFrom 3.75M credits a month, or custom
Simultaneous generations
4
Generations per minute, per key
60
Characters per hour
1,000,000
  • Up to 10,000 characters per request.
  • A failed generation is never charged.
  • Tags are not billed as text; sound effects are free too.
  • Go over a limit and you get a 429 with Retry-After and X-RateLimit-* headers.
Quickstart

Your first voice, in three steps.

  1. Create your key

    In Developers, inside your account. It starts with vf_.

    export VOICEFINDER_API_KEY="vf_…"
    Go to my keys 
  2. Install the SDK

    The official ElevenLabs one. Or use REST, nothing to install.

    pip install elevenlabs
  3. Make your first call

    The audio comes back in the response, already marked.

    curl -X POST \
      "https://api.voicefinder.io/v1/text-to-speech/f6abdab0-2524-43d1-92e2-b39486f39fbe" \
      -H "x-api-key: $VOICEFINDER_API_KEY" \
      -H "Content-Type: application/json" \
      -d '{"text": "Hello, world.", "language": "en", "output_format": "mp3"}' \
      -o hello.mp3
    Read the docs 
FAQ

Questions from people who integrate.

How do I authenticate?

With a Voicefinder key, which starts with vf_ and is created in Developers, inside your account. Send it in the x-api-key header, in Authorization: Bearer, or in xi-api-key, the one the ElevenLabs SDK uses. A missing or invalid key returns 401; an account with no balance, 402.

What errors does it return and how do I handle them?

They all share one JSON shape: error, with status and message, and detail, with a stable code to compare in your code and the request_id. On a 429 or a 503, wait for Retry-After and retry. A 502 is a failed generation and is never charged. If you contact support, send us the request-id header.

What are the limits?

They depend on the plan: 1 to 4 simultaneous generations, 10 to 60 generations per minute per key and 20,000 to 1,000,000 characters per hour. Each request takes up to 10,000 characters. The X-RateLimit-Limit, X-RateLimit-Remaining and X-RateLimit-Reset headers tell you where you stand.

Which audio formats do I get?

MP3 or WAV at 48 kHz. You can ask with the ElevenLabs names (mp3_44100_128, wav_44100…) and the x-output-format header tells you the real format delivered. PCM, μ-law and Opus are not available on Voyager 1 because they cannot carry the C2PA credential. Do not convert the file yourself: the credential is bound to its bytes; ask for the format with output_format.

Which languages does it speak?

55+, with any voice: pass the language in language or language_code. It sounds native in the languages it was recorded in and, in the rest, it may carry the accent of its main language; the x-language-warning header tells you. Spanish from Spain, Catalan, Galician and Basque, with native voices.

How do I migrate from ElevenLabs?

Change the base URL to https://api.voicefinder.io and the key to a vf_ one. The official Python and JavaScript SDK works as is. eleven_* model_id values are treated as voyager-1 with no error, and voice_id values come from our catalog: find them with voices.search().

How is it billed?

With your plan credits, from the same balance as the studio: 1 credit = 1 character. Tags are not billed as text and sound effects are free. The cost is reserved before generating and refunded in full if the audio is not delivered. The x-character-count header tells you what was charged.

Your code already speaks Voicefinder.

Get your key, change two lines and make your first call.