Introduction
Re-voice a performed narration or character line by supplying audio_url and choosing among 41 premade voices with voice_id. Alicia is the default, and remove_background_noise enables optional noise removal. Assess the returned MP3 against the source before using it in a production.
| Field | Value |
|---|---|
| Input | Audio |
| Output | Audio |
| Publisher | ElevenLabs |
| Release | November 22, 2023; introduced as speech-to-speech |
| Public API name | elevenlabs/voice-changer/audio-to-audio |
| Source input | Speech recording supplied as |
| Voice selection | 41 premade ElevenLabs voices via |
| Default voice | Alicia |
| Custom and cloned voices | Unavailable for this endpoint |
| Noise control |
|
| Converted media | MP3; same words and timing in a different voice |
| Billing unit | Source audio seconds, rounded up to whole seconds |
| Measured MP3 | 4.50 seconds; 44.1 kHz mono; 128 kbps; Brian voice |
| Measured processing | About 3.7 seconds for 4.5 seconds of English narration |
Key features & highlights
Direct the performance with your voice
The input recording provides the words and their timing. ElevenLabs says Voice Changer preserves the source intonation and pacing, allowing an acted scratch read to guide the converted speech. Record a pause or a stressed word where you need it, then assess how the chosen voice carries that delivery.
Audition 41 voices from one take
Select a premade voice with voice_id and keep the source recording fixed while comparing options. Alicia is the default, and our measured test used Brian. Custom and cloned voices are unavailable through this endpoint, so plan your narrator selection around the listed voices.
A 4.50-second MP3 from a short narration
Our English narration test ran for 4.5 seconds and converted to Brian in about 3.7 seconds. The resulting MP3 measured 4.50 seconds at 44.1 kHz, mono, and 128 kbps. A local transcription reproduced the source sentence word for word; this observation covers that specific take.
Choose whether to remove background noise
The remove_background_noise switch lets you request noise removal for the source speech. Compare a short passage with the switch enabled and disabled, then listen for clear consonants and natural pauses. Keep the setting and recording conditions consistent when auditioning voices.
Request parameters
| Parameter | Required | Description |
|---|---|---|
audio_url | Yes | Supply the URL of the speech recording you have permission to transform, with a duration of up to 5 minutes. Its length determines the billed source seconds, rounded up. at least 1 characters |
voice_id | No | Choose a listed premade voice such as Brian, or omit this field for Alicia. The selection contains 41 voices and does not accept custom or cloned voices. |
remove_background_noise | No | Set this boolean to true to request background noise removal, or false to leave that option disabled. Compare a short passage with both settings before converting a longer take. Default: |
Pricing
$0.0048 / second
Charged per second of the input recording, rounded up to the next whole second, with a 1-second minimum and a 5-minute maximum. The voice and the noise setting leave the price unchanged. Nothing is charged if the task fails.
Use cases
Replace a scratch narrator
Supply your own recorded read of a tutorial introduction, including pauses for the on-screen steps. Convert it to a selected premade voice to receive an MP3 for reviewing the narration against your edit.
Prototype game character lines
Have an actor who agreed to the transformation record a line such as “The gate is open; follow me.” Re-voice that take with a listed voice to get an MP3 that follows the actor’s timing and performed emotion.
Change an interview speaker’s voice
Use a short interview answer whose speaker has permitted voice transformation as the input. The returned MP3 presents the answer in a different premade voice; review the words and the speaker’s agreement before sharing it.
Audition a voice-over performance
Record a product walkthrough passage with the emphasis you intend to use, then submit the same excerpt with different voice selections. Compare the MP3 results to choose a narrator before processing the remaining passages.
Prompt guide & pro tips
Prepare a permitted speech recording
Record the words, pauses, and emotion you want in a take of up to 5 minutes. Accept the website likeness and consent notice before initial use, and confirm you have the right to transform the recording.
Choose a premade voice
Select
voice_idfrom the 41 available voices, or omit it to use Alicia. Decide whetherremove_background_noiseshould be enabled for your source.Call the audio-to-audio endpoint
Call elevenlabs/voice-changer/audio-to-audio with JSON containing
audio_url, your chosenvoice_id, and theremove_background_noiseboolean. The endpoint readme shows a Brian request using those fields.Review the returned MP3
Listen to the converted MP3 against the original, checking wording, pauses, and sentence endings. Use the input duration, rounded up to whole seconds, with the displayed price to understand billing.
ElevenLabs Voice Changer API: recorded speech to a new voice
ElevenLabs Voice Changer is ElevenLabs' speech-to-speech tool for re-voicing an existing performance while retaining its words and timing. Call elevenlabs/voice-changer/audio-to-audio with a speech recording of up to 5 minutes to receive an MP3 in a selected premade voice.
ElevenLabs says the source intonation and pacing carry into the converted voice. Its Voice Changer announcement describes preserving the performed delivery, so record the pauses and emphasis you intend listeners to hear.
Make an audio-to-audio request
Supply audio_url and select a premade voice_id to convert your recording. This illustrative JSON input calls for Brian without background noise removal; replace the example URL with your permitted source recording when calling elevenlabs/voice-changer/audio-to-audio.
{
"audio_url": "https://example.com/permitted-scratch-read.wav",
"voice_id": "Brian",
"remove_background_noise": false
}
- Prepare a speech take within the 5-minute input limit.
- Accept the likeness and consent notice on the website before initial use.
- Submit the JSON input to the named endpoint.
- Review the resulting MP3 against the source before using it in your edit.
Understand the response media
The endpoint's media result is converted speech in MP3 format, with the source words and timing in the selected voice. The following response example describes the measured media result, rather than an API response envelope or a promised encoding for every conversion.
Converted audio: MP3
Selected voice: Brian
Duration: 4.50 seconds
Sample rate: 44.1 kHz
Channels: mono
Bitrate: 128 kbps
Use that example to plan an output check: listen to the words, compare pause positions, and check the file's duration. Keep the source available to locate changed words or awkward transitions.
Set the three input parameters
The request uses an audio URL, a premade voice choice, and an optional noise-removal switch. There is no prompt field for rewriting the script; perform the words you want in the input recording.
| Parameter | Value or limit | Practical choice |
|---|---|---|
audio_url | Source speech URL; up to 5 minutes | Use a permitted recording with the delivery you want to retain. |
voice_id | 41 premade voices; Alicia by default | Choose a listed voice such as Brian; omit the field for Alicia. |
remove_background_noise | Boolean: true or false | Enable it to request background noise removal. |
Custom and cloned voices are unavailable for this endpoint. A personal voice identifier is therefore unsuitable even if you have used it in a different voice workflow.
Calculate duration-based billing
Billing follows source audio seconds rounded up to the next whole second. Use the current price displayed on this page when estimating a conversion; the length of the source recording is the relevant duration.
Audition an excerpt before converting the full narration. Compare that excerpt across voice choices to select a narrator before processing the remaining speech.
Prepare an effective source performance
A deliberate source read gives the conversion a clear performance to follow. Record a tutorial introduction with the pauses needed for its screen actions, or ask a consenting actor to perform a character line with the intended emotion.
- Speak the exact script, including names and sentence endings, then review it before submission.
- Keep the same source take while comparing premade voices so the performance stays fixed.
- Compare noise removal on and off using a passage that includes both speech and pauses.
- Re-record a phrase when its delivery needs changing, then convert that revised passage.
Interpret our measured result
Our test converted 4.5 seconds of English narration to Brian in about 3.7 seconds of processing. The output was a 4.50-second MP3 at 44.1 kHz, mono, and 128 kbps, and a local transcription matched the source sentence word for word.
That result supports this specific short narration, rather than guaranteeing the same processing time or word accuracy for your material. Compare quality and elapsed time on a representative passage before scheduling a longer sequence of conversions.
Resolve input and listening problems
A source longer than 5 minutes exceeds this endpoint's duration limit. Prepare separate passages within the limit and review the joins when you assemble their converted MP3s.
An unsupported voice choice should be replaced with a listed premade voice, or omitted to select Alicia. If the apparent voice or accent differs from your expectation, compare several options using the same clearly spoken passage: the source performance continues to guide delivery.
A noisy result calls for a short comparison with remove_background_noise enabled. If a word sounds wrong, check the original take, speak that phrase clearly again, and review the new output rather than treating one successful transcription test as a guarantee.
Choose conversion, translation, or a new script
Voice conversion fits an existing performed recording; translation requires a different workflow. Use the distinction below to choose what to submit next.
| Starting point and goal | Workflow to choose |
|---|---|
| A spoken take whose words and timing should remain | This Voice Changer audio-to-audio endpoint |
| Recorded speech that needs a target language | ElevenLabs Dubbing |
| A written script awaiting a spoken performance | A text-to-speech workflow |
Confirm permission before transforming a voice
The website requires a likeness and consent notice before initial use of Voice Changer. Use text and voices you have the right to use, transform your own speech or recordings you have permission to transform, and never impersonate a real person.
Use text and voices you have the right to use, and transform your own speech or recordings you have permission to transform. Never impersonate a real person with the converted audio. Accept the website likeness and consent notice before initial use.
API quick start
# Generate the idempotency key once; reuse it when retrying after a lost response.
export SPICY_IDEMPOTENCY_KEY="${SPICY_IDEMPOTENCY_KEY:-$(uuidgen)}"
curl -X POST https://api.spicyapi.ai/api/v1/jobs/createTask \
-H "Authorization: Bearer $SPICY_API_KEY" \
-H "Content-Type: application/json" \
-H "Idempotency-Key: $SPICY_IDEMPOTENCY_KEY" \
-d '{
"model": "elevenlabs/voice-changer/audio-to-audio",
"input": {
"audio_url": "https://example.com/reference.mp3",
"voice_id": "Alicia"
}
}'
# Copy data.taskId from the response into TASK_ID, then poll the status.
TASK_ID="TASK_ID_FROM_RESPONSE"
curl "https://api.spicyapi.ai/api/v1/jobs/recordInfo?taskId=$TASK_ID" \
-H "Authorization: Bearer $SPICY_API_KEY"
See the API reference for authentication, the complete schema, polling, and error handling.




