ElevenLabs Voice Changer: Transform a Recording Without Losing Its Delivery
Upload or record an authorized voice clip in ElevenLabs Voice Changer, choose a licensed target voice and suitable speech-to-speech model, then generate and check the result. Each conversion has a five-minute limit.
What is ElevenLabs Voice Changer, formerly Speech-to-Speech?
ElevenLabs Voice Changer transforms an existing spoken performance into a different available voice while aiming to retain the source speaker's cadence, emotion, pauses and pronunciation. It is a voice-to-voice conversion tool: you begin with an actual audio performance, rather than a script. ElevenLabs previously called the capability Speech-to-Speech (STS). Its documented applications include dialogue, expressive voiceovers and replacing individual spoken lines.
Suppose a video creator records a fictional character's dialogue with the intended timing, whispers and emphasis, then converts that authorized recording to a different licensed voice. The goal is to keep the acting choices while changing vocal identity. The conversion does not grant new rights to the original recording or the target voice, and it cannot promise exact preservation of every nuance. This article follows official product documentation and does not claim a hands-on test or measured audio quality.
Voice Changer vs. cloning, Voice Design and Text to Speech
Voice Changer uses an existing audio clip as the performance source. Text to Speech (TTS) starts with written text and generates spoken audio. Voice Cloning creates a reusable voice from permitted recordings, while Voice Design generates a voice concept from a description. These tools solve different problems, even when you ultimately choose the same output voice in ElevenLabs.
Pick Voice Changer if you need the pauses, intensity and expressive delivery from an actual take. Choose Text to Speech if a script is already correct and you want a fast read without recording it. Choose Voice Cloning only when you have the necessary source rights and comply with ElevenLabs' verification rules. If you need a new language instead of a new voice identity, use a translation or dubbing workflow rather than assuming Speech-to-Speech translates your dialogue.
Step 1 — Prepare a clear and legally usable source recording
Prepare speech recorded by you or obtained with the permissions necessary for the intended use. ElevenLabs' Voice Changer product guide allows MP3, M4A, FLAC, OGA, OGG and WAV audio, and lists MKV, WEBM, MP4 and MOV as supported video containers. The uploaded file must be under 50 MB and each recording or live input must be at most five minutes long. Video containers can contain more than just one clean voice track; check the audio before submitting it.
Record in a quiet room at a consistent distance from the microphone. Avoid clipping from excessive input gain and keep music below the spoken line. Before generating, listen for missing words, interruptions and overlapping speakers: the model follows the delivered performance rather than correcting a fundamentally broken take. Keep an unchanged copy of the original recording in secure project storage. Official source: https://elevenlabs.io/docs/eleven-creative/playground/voice-changer.
Step 2 — Upload audio or record directly in ElevenLabs
Open ElevenLabs Voice Changer. Drag your recorded file onto the audio area or select it with the upload control. Alternatively, choose Record Audio, activate the microphone and stop when your performance is finished. Play the recording inside the tool before selecting the destination voice. The interface shows the estimated character cost; ElevenLabs states that recording a take alone does not consume the generation quota and billing occurs when you select Generate.
For the first evaluation, use a 15- to 25-second sample that includes one short sentence, a pause and a contrasting emotional delivery. Keep this exact source clip for each output-voice comparison. Recording a new script each time would confound source quality with the target voice's contribution. This is a suggested test design rather than an experiment Claim26 has run.
Step 3 — Choose a target voice you are actually entitled to use
Choose a destination voice available to your account: for example, a licensed voice from Voice Library or a voice you have legitimately designed or cloned. Look for a suitable timbre and storytelling character, not just a popular voice. ElevenLabs documents that custom voices in your library can be addressed through their voice ID for developer integrations, but you do not need to call the API to use the web interface.
Do not treat a publicly accessible recording as permission to imitate its speaker. ElevenLabs restricts Professional Voice Cloning to the speaker's own verified voice; even consent does not permit you to create somebody else's professional clone within your own account. That speaker can instead create and verify a clone in their account and share it under the platform's rules. For a commercial project, document the recording rights, voice-access terms and any contractual restrictions. Reference: https://elevenlabs.io/docs/help-center/product/voices/voice-cloning/can-i-create-a-professional-voice-clone-of-someone-elses-voice.
Step 4 — Select the correct speech-to-speech model and language
ElevenLabs documents two speech-to-speech v2 model identifiers: eleven_english_sts_v2 for English-only conversion and eleven_multilingual_sts_v2 for its multilingual capability. The multilingual model is documented for 29 languages, including English, French, Spanish, German and Portuguese. The provider says the multilingual model can perform better even with some English material, so do not assume the English-only label always means a higher-quality result.
The source performance has a strong influence on the result. ElevenLabs illustrates this with accents: a British-sounding destination voice given American-accented source speech may retain American cadence and accent. This can be useful for consistency but undesirable if you intended a different accent. Test the same target voice with a representative source recording in the actual project language before committing to a full session. Source: https://elevenlabs.io/docs/overview/capabilities/voice-changer.
Step 5 — Control background noise and the recording's energy
Voice Changer includes a Remove Background Noise option. Turn it on when room noise or hum interferes with the source. Compare a short generated passage with and without the option, because denoising may also change the character of a recording you deliberately made in an ambient setting. Reducing noise cannot guarantee that overlapping dialogue or a badly clipped microphone recording will become intelligible.
Vary the human performance before turning to advanced settings: clear articulation, intentional pauses and moderate microphone gain will often produce a more meaningful difference than repeatedly changing the same output voice. If the actual task is removing noise while preserving the original person's voice, ElevenLabs Voice Isolator is more directly relevant than voice conversion. If your desired output needs a different language, use Dubbing rather than assuming the noise-removal checkbox translates speech.
Step 6 — Generate, listen critically and download the result
After reviewing the chosen model, source clip and target voice, select Generate. Listen first to the opening and final words, then check whether laughter, whispers, stress and pauses survived the conversion. Compare outputs at matched playback volume: a louder output may seem subjectively better even when its pronunciation is not. If a passage sounds unnatural, first revisit the source performance and target voice rather than claiming one particular parameter is universally effective.
ElevenLabs says you can download a completed conversion immediately or access Voice Changer History to retrieve earlier generated audio. The documented history interface offers MP3 or WAV and additional export options under Advanced. Keep only files you are allowed to retain and distribute. Recording a clean audition, generating a sample and exporting the final file are distinct actions with different rights and billing implications.
How much does ElevenLabs Voice Changer cost in credits?
ElevenLabs' current product guide describes Speech-to-Speech generation as 1,000 characters of quota per minute of processed input audio. For an illustrative two-minute file, that is approximately 2,000 character units; for the five-minute maximum, approximately 5,000. The calculation is based on duration, not the number of written characters in the spoken transcript. Generating multiple versions consumes additional quota according to the current plan rules.
Do not confuse the credit-equivalent calculation with a guaranteed euro or dollar cost. The provider explicitly says Voice Changer's current pricing differs between its website plans and API usage and may change. Check your available credits, the displayed generation cost and official plan terms before purchasing or running large batches. Reference: https://elevenlabs.io/docs/eleven-creative/playground/voice-changer and https://elevenlabs.io/voice-changer.
How to process an interview or audiobook longer than five minutes
ElevenLabs caps each Voice Changer conversion at five minutes regardless of subscription tier. For a twelve-minute spoken piece, prepare smaller logical chunks such as four three-minute passages or three four-minute passages. Avoid splitting in the middle of a word or breath; finish each piece at a sentence boundary and retain a reference section of the original delivery to compare across outputs.
Use the same target voice, model and settings for each chunk, then assemble the approved audio in your own audio editor. Listen around every join for abrupt loudness changes, clipped consonants and inconsistent emotion. This stitching and quality-control process is an editorial workflow; the documentation does not guarantee perfect continuity between independently generated chunks. Do not buy a more expensive tier expecting it to lift the documented five-minute per-conversion limit.
Three worked examples: character, podcast correction and multilingual narration
Example A, fictional character: a performer records the spoken scene with a whisper, a laugh and a rising question. Voice Changer can aim to preserve these choices while rendering the line with an appropriately licensed character voice. Example B, podcast repair: a presenter re-records one incorrect sentence and converts that authorized take to match an established licensed output voice. Compare timing and room sound before inserting it into the original episode.
Example C, international project: a creator records an English line, then converts it into another accessible voice while keeping English speech. Choosing a French-sounding or Spanish-sounding target does not automatically translate English to French or Spanish. Translation and language-specific dubbing require a separate process. All three examples are hypothetical illustrations, not production case studies or measured customer results.
Troubleshooting: robotic sound, wrong accent, clipping or uneven delivery
If the converted voice sounds robotic, inspect the source recording for monotone delivery and try a clearer performance; then compare two target voices with the same input. If emotion seems lost, check whether the recording actually contains the intended whisper, laugh or emphasis. If the output accent is unexpected, review the source accent and the language support of the selected model rather than assuming the destination speaker's label controls everything.
If the audio has distortion, reduce recording gain and re-record; if it contains hum, test the background-noise option. If a long recording fails, confirm both the five-minute limit and the under-50 MB file-size limit. If a segment begins or ends abruptly, split at natural sentence boundaries. For pronunciation errors repeated in synthesized scripts rather than recorded performance, a pronunciation dictionary in Text to Speech may be a more suitable solution. These are diagnostic checks, not guarantees of a repair.
Can you use a converted voice in a paid project?
Commercial rights depend on how the content was generated and whether you own or license the underlying material and target voice. ElevenLabs says its Free plan does not include a commercial-use license. Its help center describes commercial use for eligible content generated while on a paid plan, subject to the terms of service, the intellectual-property rights involved, and special exclusions such as some beta products. A paid subscription does not grant rights over somebody else's performance.
Before monetizing a course, advertisement or audiobook, check the active plan at generation time and keep evidence of the recording rights, target-voice permissions and relevant license terms. Where the platform calls for attribution, provide it as instructed. Do not publish an imitation designed to mislead listeners into believing an unrelated person endorsed your product. Reference: https://elevenlabs.io/docs/help-center/legal/can-i-publish-the-content-i-generate-on-the-platform.
A practical checklist before converting your next voice recording
First define the task: change an authorized speaker's voice without changing the language or intended performance. Second record or upload a clear excerpt under five minutes and 50 MB. Third choose a licensed destination voice and appropriate model. Fourth verify the estimated quota and optional noise reduction. Fifth run a small comparison using the same source clip, listen at the same volume, and only then generate and export longer project segments.
Keep source recordings, generated versions, voice IDs and model details with the project while respecting privacy and your licenses. If you really need narration from text, a newly designed voice, a personal voice clone, translation or noise removal, switch to the more appropriate ElevenLabs tool. Claim26 may receive an affiliate commission from qualifying subscriptions through its disclosed commercial links at no extra cost to the reader. This guide was researched from official documentation; Claim26 does not claim to have tested an ElevenLabs paid account.
Check these four signals.
✓Record or upload under five minutes and 50 MB
✓Keep original delivery and choose an authorized voice
✓Check multilingual STS model and quota before generation
✓Compare audio, verify rights and export the result
The safest way to choose is to build a small real-world workflow.
Test the part of the customer journey that matters most before moving your entire system.
Official sources for this guide
Software pricing and features can change. Review the official documentation below before choosing a product or plan.
