A voice remix should solve a casting problem you can name. If a narrator needs a slightly warmer timbre, changing age, accent, speed, and intensity at once makes the result difficult to evaluate. This ElevenLabs voice remixing guide uses a controlled audition to decide whether a change deserves to become a saved voice.
Download the voice-remix audition sheet · Audition your narration script
ElevenLabs documentation checked September 16, 2026. The audition scripts and review method are original teaching materials. No generated audio benchmark is claimed. The hero is an AI-generated illustration.
Decide whether you need a remix
ElevenLabs’ Voice Remixing documentation describes changing attributes of an existing eligible voice through a text description, previewing the result, and saving a new variant. The original voice remains available. Higher prompt strength can produce a larger departure from the source identity.
Eligibility matters. The remixing guide describes supported owned voices and eligible library voices; do not assume every voice in a catalog can be remixed. Check the current permission and availability information for the specific voice you intend to use.
Ask whether the desired change belongs to the voice or the line. “This narrator should have a softer texture across the series” suggests a reusable casting change. “Whisper this warning” may be a performance direction for one sentence. Solving the second problem by redesigning the whole voice creates unnecessary work.
Write a one-sentence casting brief
Our fictional museum audio guide needs a narrator who sounds approachable without becoming sleepy. The source voice is clear but slightly severe. The proposed change is a warmer, less metallic timbre while retaining the recognizable speaker and measured pace.
Write the invariants beside the requested change: keep the speaker’s perceived identity, accent, articulation, and approximate speaking pace. These are review goals rather than guaranteed controls. The preview still has to demonstrate them.
This brief deliberately avoids a stack of dramatic adjectives. If you need a different accent and a different character, write that as a different casting task and evaluate it separately. A narrowly defined audition gives you a clearer stopping point.
Use an audition script that exposes weaknesses
A single enthusiastic sentence may flatter almost any voice. Use a short passage with a statement, a question, a number, a proper name, and a quieter ending. Keep the exact same text for every candidate so changes in wording do not influence the comparison.
The fictional passage tests clarity, pronunciation, question intonation, and a restrained invitation. It does not prove the voice works for every language or genre. Replace names and numbers with the ones your real project actually contains.
Generate or retain a baseline reading from the source voice. Label files neutrally, such as A, B, and C, when asking another person to listen. Descriptive filenames such as “warmer-best-final” can bias the judgment before playback begins.
Compare a small range of strengths
Start with a modest prompt strength and preview the result. If the change is too subtle, try the next suitable level while keeping the brief and text fixed. Do not jump to the strongest setting simply because the interface offers it.
ElevenLabs documents strength levels and notes that stronger changes can alter identity more substantially. Treat that as a tradeoff to hear, not a reason to assume that maximum strength produces maximum quality. Review any displayed generation charges before expanding the audition.
| Candidate | Question | Record |
|---|---|---|
| Source | What problem does the original actually have? | A specific word or passage where severity is distracting |
| Modest change | Is the warmth audible without identity drift? | Target improvement and any pronunciation change |
| Stronger change | Does the added change still fit the same narrator? | Identity, pace, and texture differences |
Keep playback levels reasonably comparable. A louder clip can seem fuller or more confident even when the underlying voice is not better. Judge the unprocessed render first; heavy music and effects can conceal weaknesses you need to hear.
Use a held-out script before saving the voice
A candidate can fit the audition passage and fail on a different sentence structure. Test the leading variant on fresh text that was not used to choose it. This is a practical check against selecting a voice solely because of one lucky preview.
This second passage asks for clear instructions rather than reflective storytelling. Listen for an overly soft warning, swallowed consonants, an unexpected accent shift, or a long pause that makes the direction confusing.
Time both passages in the actual project. If the remix consistently runs longer, decide whether the edit can accommodate it. Do not immediately speed it up to match the baseline; first establish whether the natural performance is appropriate.
Score the change against the brief
Use separate judgments for target change, speaker identity, intelligibility, and editorial fit. A simple pass, revise, or reject decision is enough. An elaborate numerical score is not more scientific if the criteria are vague.
- Target change: the voice sounds warmer in ordinary sentences.
- Identity: listeners still recognize the intended narrator.
- Clarity: the name, year, and directions remain understandable.
- Timing: both passages fit the available space without rushed endings.
- Consistency: the held-out script supports the same casting decision.
If the candidate improves warmth but fails clarity, revise the brief to preserve articulation and compare again. If it requires extensive repair on every line, the source or a different casting choice may be more suitable. Saving a voice is a convenience, not evidence that the audition succeeded.
Save a variant with enough context to reuse it
Once the candidate passes, save it through the current ElevenLabs workflow. Record the source voice, intended use, remix description, chosen settings, and the audition files. Use a descriptive production name such as “North Gallery — warm narration” rather than a vague sequence of final-version labels.
Keep project permissions and voice-use conditions with the production record. When working with a real person’s voice, use appropriate permission for the intended use. A technical remix option does not itself establish that permission.
Our voice design guide addresses starting from a new voice description. For a reusable script before external casting, audition the narration text in Voice Lab. QuestStudio can help test the writing; the eligible-voice remix and saved variant described here happen separately in ElevenLabs.
Frequently asked questions
Does remixing replace the original voice?
The documented workflow creates a new variant while preserving the original voice. Keep your baseline audition so you can compare them.
Can I remix every library voice?
No. Eligibility depends on the voice and current permissions. Check the selected voice against ElevenLabs’ current remixing documentation.
Why use a second audition script?
It checks whether the chosen voice works beyond the passage used to select it. This can reveal clarity, timing, or identity problems hidden by the first preview.
Is a remix necessary for one emotional sentence?
Often a line-level direction is the more direct choice. Use a remix when the voice change should persist across future material.