Voice remixing

Overview

ElevenLabs voice remixing is available on the core platform and via API. This feature transforms existing voices by allowing you to modify their core attributes while maintaining the unique characteristics that make them recognizable. This is particularly useful for adapting voices to different contexts, creating variations for different characters, or improving and/or changing the audio quality of existing voice profiles.

As an example, here is an original voice:

And here is a remixed version, switching to a San Francisco accent:

Usage

The voice remixing model allows you to iteratively transform voices you own by adjusting multiple attributes through natural language prompts and customizable settings.

Key Features

• Attribute Modification: Change gender, accent, speaking style, pacing, and audio quality of any voice you own
• Iterative Editing: Continue refining voices based on previously remixed versions
• Script Flexibility: Use default scripts or input custom scripts with v3 model audio tags like [laughs] or [whispers]
• Prompt Strength Control: Adjust remix intensity from low to high for precise control over transformations

Remixing parameters

Prompt Strength

Voice remixing offers varying degrees of prompt strength to control how much your voice transforms:

• Low: Subtle changes that maintain most of the original voice characteristics
• Medium: Balanced transformation that modifies key attributes while preserving voice identity
• High: Strong adherence to remix prompt, may significantly change the tonality of the original voice
• Max: A full transformation of the voice, but at the cost of changing the voice entirely

Script Options

• Default Scripts: Pre-configured scripts optimized for voice remixing
• Custom Scripts: Input your own text with support for v3 model audio tags such as:
  • [laughs] - Add laughter
  • [whispers] - Convert to whispered speech
  • [sighs] - Add sighing
  • Additional emotion and style tags supported which can help craft the voice

Tips and Tricks

Getting Started

Start with a high prompt strength early in your experimentation to understand the full range of transformation possibilities. You’ll need to have a voice to start with, if you haven’t already created a voice, experiment with default voices available in your library to understand how different base voices respond to remixing.

You can create custom voices using Voice Design as starting points for unique remixes.

Advanced Techniques

• Iterative refinement: Sometimes multiple iterations are needed to achieve the desired voice quality. Each remix can serve as the base for the next transformation
• Combine attributes gradually: When making multiple changes (e.g., accent and pacing), consider applying them in separate iterations for more control
• Test with varied content: Different scripts may highlight different aspects of your remixed voice

Supported Voice Formats

Input

• Any cloned voice that you personally own (Instant Voice Clone or Professional Voice Clone)
• Voices created through our Voice Design product

Output

• Full-quality voice model in v3 (but backwards compatibility to all other models)
• Iteratively editable voice that can be further remixed

FAQ