The work is executed in the page through browser workers, WebAssembly, Web Audio, or WebGPU as stated.
CLONE / AVAILABLE NOW
Voice Cloning
Authorized voice matching on your GPU. Generate a short, watermarked speech sample from an authorized reference voice using local WebGPU inference.
CLONE / local workbenchDesktop WebGPU · first model load ≈1.5 GB
GUIDE / voice cloning online
What does voice cloning do?
Voice cloning estimates a speaker representation from a reference recording and conditions generated speech on that representation. SoundTools follows Resemble AI’s official browser architecture: four Chatterbox ONNX sessions load on demand in a Web Worker, the local reference is encoded once per session, and a short watermarked waveform is returned to the page.
RELATED USES
- local AI voice clone
- Chatterbox voice cloning
- clone my voice browser
- WebGPU voice cloning
FIELD NOTES / 01
A real local processor, not a decorative upload box
Chatterbox ONNX through Transformers.js WebGPU runs after you choose a source and explicitly start the operation. The audio samples remain in browser memory; SoundTools does not send the selected media to an application endpoint.
Confirm voice authorization, provide a clean 5–10 second reference, download the approximately 1.5 GB model once, encode the speaker, enter no more than 180 characters, and adjust expressiveness from 0 to 1.5. Results remain available for preview and download only in the current tab. Closing or refreshing the page releases those temporary objects.
PROCESS / THREE STEPS
How to use voice cloning
- 01
Choose the local source
Open a compatible file in the dedicated workbench at the top of this page. The browser validates and decodes it without an upload step.
- 02
Set the useful controls
Confirm voice authorization, provide a clean 5–10 second reference, download the approximately 1.5 GB model once, encode the speaker, enter no more than 180 characters, and adjust expressiveness from 0 to 1.5.
- 03
Process, check, and export
Run the local processor, inspect the status and preview, then save Short watermarked 24 kHz WAV when the result is ready.
SIGNAL NOTES / VERIFIED BEHAVIOR
What this tool actually does
Clear limits are part of a useful tool. These values describe the processor currently running in this page.
Runtime and model assets may be downloaded, but the selected file is not attached to those requests.
The dedicated workbench exposes only formats it can actually produce in the current browser.
USE CASES / 02
Useful reasons to open voice cloning
- Your own accessibility voice
Test a short authorized phrase in a voice sample you control.
- Consented character prototype
Preview a line for a performer who has explicitly approved the use.
- Private browser experiment
Evaluate zero-shot cloning without sending the reference to an inference API.
QUESTIONS / PRACTICAL ANSWERS
Questions about this tool
Answers based on the current browser processor—not promises about a future version.
01Is my audio uploaded for processing?
No. Audio samples stay in browser memory. The page may download a codec or model asset, but it does not send the selected file with that request.
02Why can the first run take longer?
This tool uses Chatterbox ONNX through Transformers.js WebGPU. A browser may need to download and initialize that runtime before the first operation; later runs can reuse cached assets.
03Can I preview the result before saving?
Yes. Audio-producing tools expose local result players, while analysis and transcription tools show their detected data before download.
04What happens when the browser is unsupported?
The workbench reports the missing capability instead of uploading the media or pretending that processing is still running.
05Do I need an account or install?
No. The tool opens directly in a compatible browser, requires no SoundTools account, and creates its temporary working state on the current device.
06Can I clone another person without asking?
No. The workbench requires you to confirm ownership or explicit permission and must not be used for impersonation, fraud, harassment, or deception.
07Why is the first download so large?
Chatterbox uses four ONNX sessions totaling about 1.5 GB. Completed assets can be reused from browser cache on later visits.
08Is generated audio watermarked?
Yes. SoundTools keeps the Chatterbox generation path and its built-in provenance watermarking intact.