Chatterbox TTS Review
github.com · checked 21 Oct 2026
Open-source text-to-speech and voice-cloning models that run locally or through an API.
How it plugs in
Signal on Image Upscaling Software7.6#15 of 99
Chatterbox TTS is Resemble AI’s open-source family of text-to-speech and voice-cloning models. It runs on Linux, macOS, Windows, and self-hosted setups, with CUDA, CPU, or Apple MPS device choices. Models include English, multilingual, single-language, low-latency, and CPU-focused options. Audio exports as WAV and includes a PerTh watermark.

Review
In a line
Chatterbox TTS is a family of open-source models for turning text into speech and cloning voices. The project is made by Resemble AI.
What it does
You can generate speech from text, clone a voice, and save the result as a WAV file. The model range includes Chatterbox Turbo, Nano, Multilingual V3, and dedicated single-language models.
Turbo is a 350-million-parameter English model for lower compute and latency. It accepts tags such as [laugh], [cough], [chuckle], [sigh], and [gasp] in prompts. Nano has 110 million parameters and targets CPU or on-device use; the project claims three-times-realtime speed on eight CPU cores.
Multilingual V3 supports 23 languages, including Arabic, Chinese, English, French, German, Hindi, Japanese, Korean, Spanish, and Turkish. Dedicated 500-million-parameter packs cover Chinese, several Spanish and Portuguese variants, and Hindi.
How it connects
The code is open source under the MIT license. You can install it with pip or from source, run it locally without an account, and choose CUDA, CPU, or Apple MPS devices. It runs on Linux, macOS, Windows, and self-hosted systems.
Replicate offers an official hosted Chatterbox model with inputs for text prompts, reference audio, temperature, CFG, and exaggeration. The Python interface also exposes exaggeration, CFG weight, temperature, and random-seed controls.
Chatterbox fits projects in AI voice cloning software and voice cloning software. Generated audio is exported as WAV and carries Resemble AI’s PerTh watermark.
Good fit
Turbo suits low-latency voice agents. Nano is aimed at on-device or CPU inference when memory and response-time limits matter. V3 is the newest general-purpose multilingual model and is described as improving speaker identity and accent preservation across languages.
The project supports commercial use, while its models are also offered free for research and development.
Watch for
Running the models yourself requires setup with Python and a supported device. NVIDIA’s plugin documentation specifies 64-bit Windows 10 or 11, 16 GB of RAM, and an NVIDIA GPU with compute capability 7.0 or newer for that integration. It lists about 2.3 GB of VRAM for Turbo-only inference and 3.5 GB for multilingual TTS-only inference.
Every generated audio file includes a PerTh watermark. Resemble AI says it has stopped offering voice generation as a commercial product, so hosted commercial availability is not the same as the open-source license.
Pricing
The open-source models are free under the MIT license, and the project lists a free Voice AI Research plan. Self-hosting does not require an account.
Replicate’s hosted Chatterbox model costs $0.025 per 1,000 input characters.
Verdict
Choose Chatterbox if you want free, open code for speech generation or voice cloning and can manage installation or a hosted API. Its range covers multilingual output, expressive tags, low-latency English, and CPU-focused use.
Look elsewhere if you need audio without the PerTh watermark, a commercial hosted voice service from Resemble AI, or a turnkey app instead of model setup.
Our take
Chatterbox TTS is a practical option for developers and teams that can run open-source models or pay for the Replicate API. Its model choices cover multilingual speech, expressive English prompts, and smaller CPU deployments, while the MIT license allows commercial use. The trade-offs are setup work, hardware requirements for some integrations, WAV-only export, and a watermark on every generated file. Resemble AI’s withdrawal from commercial voice generation also matters if you need an ongoing hosted service from the maker.
- Free open-source models under the MIT license
- Voice cloning and text-to-speech in one model family
- 23 languages in the Multilingual V3 model
- Turbo supports expressive tags such as [laugh] and [cough]
- Runs with CUDA, CPU, or Apple MPS
- Every generated audio file includes a PerTh watermark
- Self-hosting requires Python and device setup
- WAV is the listed export format
- Resemble AI stopped commercial voice generation
- Some NVIDIA integrations require 16 GB RAM and a supported GPU
- Pick it ifTeams building speech or voice-cloning features; Local, self-hosted, or CPU-focused deployments; Projects needing multilingual output
- Skip it ifYou need unwatermarked audio; You want a turnkey hosted service from Resemble AI; You cannot manage local model or GPU setup
Pricing
The open-source models are free under the MIT license. Replicate hosts Chatterbox for $0.025 per 1,000 input characters.
- Chatterbox open-source modelsFree to use under the MIT license · Open-source models for research and developmentFree
- Voice AI Research (Chatterbox)Open models · free to build on · no commercial hosted voice serviceFree
As github.com lists them, checked 8 Oct 2026.
Questions
Is Chatterbox TTS free?
Yes. The open-source models are free under the MIT license. Replicate’s hosted version charges $0.025 per 1,000 input characters.
What languages does Chatterbox support?
Chatterbox Multilingual V3 lists 23 languages, including Arabic, Chinese, English, French, German, Hindi, Japanese, Korean, Spanish, and Turkish.
Can Chatterbox clone voices?
Yes. Chatterbox is an open-source family for both text-to-speech and voice cloning.
Where can Chatterbox run?
It supports Linux, macOS, Windows, self-hosted setups, and API use. Device examples include CUDA, CPU, and Apple MPS.
Does Chatterbox add a watermark?
Yes. The README says every generated audio file includes Resemble AI’s PerTh watermark.
What file type does it export?
WAV is the listed audio export format.
Does Chatterbox have an API?
Replicate provides an official hosted Chatterbox model with prompt, reference-audio, temperature, CFG, and exaggeration inputs.
Near it
All alternatives →- MI
Media.io AI Image Upscaler#14 on this board - N
NextGenUp#16 on this board - P
Pixelcut#13 on this board - PI
PicWish Image Upscaler#17 on this board - TG
Topaz Gigapixel#12 on this board - XIXimilar Image Upscaler#18 on this board
- PAPhot.AI#11 on this board
- C
Claid#19 on this board - CPCutout.Pro#10 on this board
- ISIcons8 Smart Upscaler#20 on this board