chatterbox-turbo-GGUF Text-to-Speech Model: 4GB+ VRAM, File List

At a Glance
| Item | Value |
|---|---|
| Repository | ggml-org/chatterbox-turbo-GGUF |
| Published | 2026-11-06 |
| License | mit |
| Formats | GGUF |
| Source type | Primary source (the publisher itself) |
Values determined by this site’s code when the information was collected. Dates are JST.
Overview
ggml-org has released “ggml-org/chatterbox-turbo-GGUF," a quantized model converted to GGUF format for “Chatterbox-Turbo," an open-source text-to-speech (TTS) model developed by Resemble AI.
This model is a Text-to-Speech model that generates high-quality audio based on input text. It can be run directly on local PCs or servers by utilizing the llama-tts command-line tool included in the llama.cpp project or the /tts endpoint of llama serve.
Specifications
The main specifications for the original model ResembleAI/chatterbox-turbo and this GGUF-converted model are as follows:
- Parameter Count: 350M (based on the original Chatterbox-Turbo specifications)
- Supported Languages: English-only (as it is an English-only variant, the language option
--tts-langis not used) - Voice Cloning: Supports zero-shot voice cloning by specifying a reference audio file (wav or mp3 format) via
--tts-speaker-file(if not specified, the model’s built-in default voice is used) - Paralinguistic Tag Support: Natural nuances such as laughter or coughing can be generated by including tags like
[laugh],[cough], or[chuckle]in the text - Decoder Technology: The decoder portion from audio tokens to mel-spectrograms has been distilled, reducing the number of steps required for audio generation from 10 steps to 1 step
- Watermarking: All generated audio includes Resemble AI’s “PerTh (Perceptual Threshold)" neural watermark, which is designed to be detectable even after editing or MP3 compression
- License: MIT License
Performance and Quality
A comparison of the “Chatterbox" family and the model architecture for this GGUF version is shown in the table below, as provided in the official model card.
→ Scroll horizontally to see all columns
| Model | Size | Languages | Key Features | Best For | 🤗 | Examples |
|---|---|---|---|---|---|---|
| Chatterbox-Turbo | 350M | English | Paralinguistic Tags ([laugh]), Lower Compute and VRAM |
Zero-shot voice agents, Production | Demo | Listen |
| Chatterbox-Multilingual (Language list) | 500M | 23+ | Zero-shot cloning, Multiple Languages | Global applications, Localization | Demo | Listen |
| Chatterbox (Tips and Tricks) | 500M | English | CFG & Exaggeration tuning | General zero-shot TTS with creative controls | Demo | Listen |
In performance comparisons of the original models, Chatterbox-Turbo is lighter at 350M parameters compared to the 500M parameters of the existing Chatterbox or Chatterbox-Multilingual, making it superior for resource-efficient operation with lower compute and VRAM usage. Additionally, because it achieves high-speed generation in a single step due to decoder distillation, it exhibits excellent low-latency characteristics for voice agent applications where real-time performance is required. On the other hand, since the supported language is limited to English only, there is a trade-off where it cannot be used for multi-language synthesis like the Chatterbox-Multilingual, which supports over 23 languages.
Strengths and Use Cases
Here is an explanation of the strengths and intended use cases based on the design and functionality of the original “Chatterbox-Turbo" model. This model excels in the following scenarios and applications:
-
Low-Latency Voice Agents and Conversational Systems:
The number of generation steps has been reduced to one through decoder distillation, keeping processing latency extremely low. Therefore, it is suitable for conversational AI agents where response speed is prioritized and for operation in production environments that require real-time performance. -
Synthesis of Conversational Audio with Rich Emotion and Realism:
By including paralinguistic tags such as[cough],[laugh], and[chuckle]in the text, you can create expressive audio that differs from conventional mechanical speech synthesis. This is effective for content creation and narration where natural nuances of conversation are required. -
Zero-Shot Voice Cloning:
You can read text while reproducing a speaker’s voice without additional fine-tuning simply by specifying a short reference audio file (wav or mp3 format). You can immediately realize speech in the voice of a specific character or person. -
English Audio Content Production:
As it is tuned as an English-only variant, it can generate high-quality audio for reading English text.
Hardware Requirements
Estimated requirements (calculated by Local Model Watch)
| Your VRAM | Quantization | File size | Est. memory needed |
|---|---|---|---|
| 4GB (laptop iGPU / phone class) | BF16 | 0.8GB | 1.0GB |
Memory estimates add a 20% runtime overhead (KV cache, etc.) to the actual size of the distributed files. Actual usage varies with context length, batch size and inference engine. These figures are computed by this site from file sizes, not published by the model’s authors. Compare with other models in our VRAM quick reference. What the quantization names mean: glossary.
Can You Run It Locally?
The publisher distributes this model as GGUF.
License — mit (Commercial use allowed): Permits commercial use, modification and redistribution, provided the copyright notice and license text are retained.
Compiled by this site’s code from the published formats and the license field. License summaries are not legal advice — check the publisher’s original terms before relying on them.
Distributed Files
Weight files published in ggml-org/chatterbox-turbo-GGUF, listed by this site from the Hugging Face API. Sizes are the actual file sizes.
| File | Size |
|---|---|
chatterbox-turbo-BF16.gguf |
874MB |
chatterbox-turbo-Q8_0.gguf |
482MB |
mmproj-chatterbox-turbo-F16.gguf |
555MB |
How to Get It
“ggml-org/chatterbox-turbo-GGUF" can be obtained from the Hugging Face model repository. This model is not gated, meaning you do not need to agree to terms of service to download and use it directly.
It is provided in GGUF format and can be run on local environments or servers using tools integrated into the llama.cpp project. The main usage and command procedures are as follows:
-
Direct Execution from Terminal (llama-tts):
Use the command-line toolllama-ttsto generate an audio file (WAV format) from text.bash llama-tts -hf ggml-org/chatterbox-turbo-GGUF -p "Hello world" --output out.wav -
Execution in Server Mode (llama serve):
You can startllama serveto provide audio synthesis functionality via the/ttsendpoint.bash llama serve -hf ggml-org/chatterbox-turbo-GGUF -
Specifying Voice Cloning:
If you want to reproduce a voice using a reference audio file (wav, mp3), add the--tts-speaker-fileoption and specify the path to the audio file. If not specified, the default voice built into the model is used. Note that since this model is English-only, the language option--tts-langis not used.
For detailed usage instructions and tool implementation specifications, you can also check the llama.cpp Chatterbox documentation.
Related Articles
- higgs-audio-v3-stt-GGUF Speech-to-Text Model: 4GB+ VRAM, File List
- Soprano-1.1-80M-GGUF Lightweight English TTS Model: File List
- chatterbox-GGUF 23-Language Multilingual TTS Model: 4GB+ VRAM
- Voxtral-Mini-4B-Realtime-2602-GGUF Real-time ASR Model: 4GB+ VRAM
What to Read Next
- Find models by VRAM (This model runs from the 4GB tier) → Other models that run on a 8GB GPU
- Formats this model is available in → GGUF format guide and models
- Other models for the same task → Other audio models

