chatterbox-tts

What it is and what it does

Chatterbox TTS is a family of neural text-to-speech models from Resemble AI that convert written text into natural-sounding speech. The package includes three model variants: Turbo (350M parameters, English-only, optimized for low-latency voice agents), Multilingual (500M parameters, supports 23+ languages), and the original Chatterbox (500M parameters, English with creative control tuning). All models support zero-shot voice cloning—you provide a reference audio clip and the model adapts its output to match that speaker's voice characteristics.

The package depends on a substantial ML stack: PyTorch, librosa for audio processing, transformers for language understanding, diffusers for generative modeling, and several specialized libraries (conformer, spacy-pkuseg for CJK text, pykakasi for Japanese). Every generated audio file includes an imperceptible neural watermark (Perth) for responsible AI tracking. The Turbo variant adds native support for paralinguistic tags like [cough] and [laugh] to inject realism, and reduces mel-spectrogram generation from 10 steps to one, trading some flexibility for speed. Configuration options (cfg_weight, exaggeration) allow tuning expressiveness and pacing.

Use it for:

• Build low-latency voice agents or conversational AI that respond with natural speech in real time.
• Generate multilingual narration for video, podcasts, or interactive media in 23+ languages with consistent voice.
• Clone a specific speaker's voice from a short reference clip for personalized TTS without retraining.
• Add expressive speech effects (laughter, coughing) to game dialogue, audiobooks, or creative projects.
• Prototype TTS features before committing to a commercial service, using the open-source models locally.

Worth the install?

AI-flagged interpretation of the facts on this page — verify before relying

Chatterbox TTS converts text to speech using open-source neural models, supporting English and 23+ languages with optional voice cloning from reference audio clips.

Yes, if you need open-source neural TTS with voice cloning and have GPU resources available. The package is actively maintained, permissively licensed, and offers competitive model variants. Install with caution on resource-constrained systems—the dependency stack is heavy. For production voice agents requiring sub-200ms latency at scale, Resemble AI's commercial service is recommended as an alternative.

Install

chatterbox-tts on PyPI

pip install chatterbox-tts

uv add chatterbox-tts

poetry add chatterbox-tts

Installing chatterbox-tts

Before you install

Low install friction with a pure Python wheel. The package is actively maintained with recent commits and substantial community interest (25990 GitHub stars). However, it carries 15 runtime dependencies including torch, torchaudio, transformers, and diffusers—a heavy ML stack that will require significant disk and memory resources during installation.

License in practice

MIT License permits free use, modification, and distribution with minimal restrictions. You may use this in commercial projects and modify the code as needed, provided you retain the copyright notice and license text.

Quickstart

pip install chatterbox-tts

import torchaudio as ta
from chatterbox.tts_turbo import ChatterboxTurboTTS

model = ChatterboxTurboTTS.from_pretrained(device="cuda")
wav = model.generate("Hello world", audio_prompt_path="ref.wav")
ta.save("output.wav", wav, model.sr)

Requires Python >=3.10. GPU (CUDA) strongly recommended for practical inference speed. A reference audio clip (approximately 10 seconds) is needed for voice cloning. The full dependency stack (torch, transformers, diffusers) will consume several gigabytes of disk space and VRAM.

Verify before relying

• Whether the Turbo model's single-step mel decoder actually maintains audio fidelity compared to multi-step alternatives in blind listening tests.
• Actual inference latency and VRAM requirements for the Turbo model on consumer GPUs.
• Whether paralinguistic tags ([cough], [laugh], etc.) work reliably across all supported languages or only English.
• Compatibility and performance on non-CUDA devices (CPU, AMD ROCm, Apple Metal).

Package facts

Evidence: chatterbox_tts-0.1.7-py3-none-any.whl

Tags

More Artificial Intelligence packages