What for: open TTS from Resemble AI with zero-shot voice cloning from a short sample — runs locally, offline, with nothing to pay.

What it is and who it's for

Chatterbox is an open TTS model from Resemble AI with voice cloning from a short sample, running locally and for free. For developers and content makers who need their own (or a specific) voice without cloud bills and without sending data anywhere. What sets it apart: zero-shot cloning in open source — the thing clouds usually charge for.

Key features

Zero-shot voice cloning from a few seconds of reference audio.
Open weights — local, offline, no fees.
Control over expressiveness and emotion in the synthesis.
A Python package, runs on GPU.
Optional watermarking of generated audio from Resemble.
Fits dubbing pipelines and character voiceover.

Get started in 5 minutes

Install it:

pip install chatterbox-tts

Synthesize with cloning from a sample:

from chatterbox.tts import ChatterboxTTS; import torchaudio
model = ChatterboxTTS.from_pretrained(device="cuda")
wav = model.generate("Привет, мир", audio_prompt_path="voice_sample.wav")
torchaudio.save("out.wav", wav, model.sr)

Swap in your own voice_sample.wav to clone that voice.

When to take it, and when not

✅ Take it if: you need free local voice cloning from a sample with control over emotion.
✅ Take it if: privacy and offline use matter and cloud TTS is off the table.
❌ Skip it → "Kokoro TTS" is better if you don't need cloning — Kokoro is lighter and simpler for ready-made voices.
❌ Skip it → "ElevenLabs" is better if you want a mature cloud dubbing ecosystem and the widest language coverage out of the box.

The honest price

Free (open-source). Your cost is a GPU for comfortable speed. For commercial use, check the license in the repo. Exact requirements depend on your hardware.

Gotchas

Cloning someone else's voice without consent is a legal and ethical risk; keep the audio watermarking in mind.
Clone quality depends on how clean the reference audio is (noise ruins the result).
Decent speed needs a GPU; on CPU it's slow.
Language support is limited — test Russian on your own samples.

🤖 Prompt accelerator

"I want to clone a voice locally and read a Russian text with Chatterbox (Resemble AI). Help me install it, prepare a clean reference sample (length, format, how to remove noise), write a generation script with audio_prompt_path, and tune expressiveness. Tell me the GPU requirements and how to stay on the right side of ethics and the license when using a voice."