Voice Changer

Voice Changer is speech-to-speech: it re-records a source file in a different voice while preserving delivery, pacing, and timing. It is asynchronous and uses its own endpoints — not /audio/queue, and not text-to-speech or voice cloning. Choose a voice-changer model, request a price quote, queue the conversion, then poll until Venice returns the converted audio.

A queued conversion is charged immediately. If the queue response is lost, poll /audio/voice-changer/retrieve with the same queue_id. Do not queue the same recording again.

Choose a model

Voice-changer models are returned by GET /models?type=music with model_spec.voice_changer set to true. There is no ?type=voice-changer filter. The examples below use elevenlabs-voice-changer.

curl "https://api.venice.ai/api/v1/models?type=music" \
  -H "Authorization: Bearer $VENICE_API_KEY"

Check each model’s metadata before setting optional fields:

Field	Use it for
`voices` / `default_voice`	Target voice names. Omit `voice` to use the default.
`supports_custom_voice_id`	Whether `voice` also accepts a provider Voice ID
`accepted_audio_formats`	Source containers Venice accepts (validated from the file’s binary signature, not the filename)
`max_source_audio_duration_seconds`	Longest source recording the model accepts
`supports_background_noise_removal`	Whether `remove_background_noise` is accepted
`supports_seed`	Whether `seed` is accepted
`pricing.durations`	Whole-minute price tiers

Unsupported fields cause an HTTP 400 response. Recordings longer than max_source_audio_duration_seconds are rejected with HTTP 422 before any charge.

Conversion flow

Endpoint	Purpose
`POST /audio/voice-changer/quote`	Estimate the conversion cost in USD
`POST /audio/voice-changer/queue`	Start a speech-to-speech conversion
`POST /audio/voice-changer/retrieve`	Poll the job and download converted audio
`POST /audio/voice-changer/complete`	Delete stored media after downloading it

1. Get a price quote

Voice Changer is billed from the length of the source recording, rounded up to the next whole minute. Quote the length you expect to send; the charge is computed from the length Venice measures when the recording is queued.

curl https://api.venice.ai/api/v1/audio/voice-changer/quote \
  -H "Content-Type: application/json" \
  -d '{
    "model": "elevenlabs-voice-changer",
    "duration_seconds": 60
  }'

The response contains the estimated cost in USD and the duration the quote was computed for:

{
  "quote": 0.35,
  "duration_seconds": 60
}

2. Queue the conversion

Supply the source recording exactly one of two ways: as a multipart file upload, or as an audio_url in a JSON body. Supplying both, or neither, is rejected. When you pass a URL, Venice fetches and validates the bytes itself and forwards only those bytes to the provider — the URL is never handed onward.

curl https://api.venice.ai/api/v1/audio/voice-changer/queue \
  -H "Authorization: Bearer $VENICE_API_KEY" \
  -F "model=elevenlabs-voice-changer" \
  -F "voice=Aria" \
  -F "file=@./source-recording.mp3"

curl https://api.venice.ai/api/v1/audio/voice-changer/queue \
  -H "Authorization: Bearer $VENICE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "elevenlabs-voice-changer",
    "voice": "Aria",
    "audio_url": "https://example.com/source-recording.mp3"
  }'

Optional fields, when the model reports support:

remove_background_noise — strip background noise before conversion
seed — integer ≥ 0 for a reproducible result

A successful request returns the model, a queue ID, and the measured source length:

{
  "model": "elevenlabs-voice-changer",
  "queue_id": "0190f2c4-9c1e-7a3b-8f42-2c9d5e7a1b34",
  "status": "QUEUED",
  "duration_seconds": 52
}

Save model and queue_id; the retrieve and complete endpoints require them. Compare duration_seconds to your quote if you need to reconcile the estimate against the billed length.

Queue is not safe to retry. A successful queue request has already been charged.

3. Poll and download

Call /audio/voice-changer/retrieve with the values from the queue response:

curl https://api.venice.ai/api/v1/audio/voice-changer/retrieve \
  -H "Authorization: Bearer $VENICE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "elevenlabs-voice-changer",
    "queue_id": "0190f2c4-9c1e-7a3b-8f42-2c9d5e7a1b34"
  }' \
  --output response.bin

Inspect the response Content-Type:

Content-Type	Meaning	Action
`application/json`	Conversion is still processing	Read the timing fields, wait, and poll again
`audio/mpeg`	Conversion is complete	Save the binary body as an `.mp3`

A processing response looks like this:

{
  "status": "PROCESSING",
  "average_execution_time": 10000,
  "execution_duration": 4200
}

Both timing values are milliseconds. A completed response also includes x-venice-audio-format, x-venice-audio-duration, x-venice-inference-time, x-venice-model-id, and x-venice-model-name. If the provider fails the conversion, the charge is refunded automatically and the error body includes credits_refunded. Polling again replays the same result rather than refunding twice. To delete stored media in the same call that returns the audio, set delete_media_on_completion to true on retrieve. The audio cannot be retrieved again afterwards.

Complete example

This Python example quotes a conversion, uploads a source file, polls every five seconds, and saves the result as MP3.

import os
import time
from pathlib import Path

import requests

BASE_URL = "https://api.venice.ai/api/v1"
HEADERS = {
    "Authorization": f"Bearer {os.environ['VENICE_API_KEY']}",
}

source = Path("source-recording.mp3")

quote = requests.post(f"{BASE_URL}/audio/voice-changer/quote", json={
    "model": "elevenlabs-voice-changer",
    "duration_seconds": 60,
})
quote.raise_for_status()
print(f"Estimated cost: ${quote.json()['quote']:.2f}")

with source.open("rb") as audio:
    queued = requests.post(
        f"{BASE_URL}/audio/voice-changer/queue",
        headers=HEADERS,
        data={
            "model": "elevenlabs-voice-changer",
            "voice": "Aria",
        },
        files={"file": audio},
    )
queued.raise_for_status()
job = queued.json()
print(f"Queued {job['queue_id']} ({job['duration_seconds']}s billed)")

while True:
    result = requests.post(
        f"{BASE_URL}/audio/voice-changer/retrieve",
        headers={**HEADERS, "Content-Type": "application/json"},
        json={"model": job["model"], "queue_id": job["queue_id"]},
    )
    result.raise_for_status()
    content_type = result.headers.get("Content-Type", "").split(";")[0]

    if content_type == "audio/mpeg":
        output = Path("converted-audio.mp3")
        output.write_bytes(result.content)
        print(f"Saved {output}")
        break

    status = result.json()
    print(f"Status: {status['status']}")
    time.sleep(5)

requests.post(
    f"{BASE_URL}/audio/voice-changer/complete",
    headers={**HEADERS, "Content-Type": "application/json"},
    json={"model": job["model"], "queue_id": job["queue_id"]},
).raise_for_status()

The quote endpoint does not require authentication, but queue, retrieve, and complete requests do.

Docs

Getting Started

Text & Chat

Image

Video

Audio

Search & RAG

Agents & Integrations

Coding Tools

Agent Tooling

SDKs & Frameworks

Choose a model

Conversion flow

1. Get a price quote

2. Queue the conversion

3. Poll and download

Complete example

​Choose a model

​Conversion flow

​1. Get a price quote

​2. Queue the conversion

​3. Poll and download

​Complete example

​Related resources

Choose a model

Conversion flow

1. Get a price quote

2. Queue the conversion

3. Poll and download

Complete example

Related resources