Chatterbox TTS — Resemble AI Open-Source TTS

Chatterbox TTS

Chatterbox is Resemble AI's SOTA open-source text-to-speech model. It outperforms ElevenLabs on standard TTS benchmarks and ships with voice cloning (zero-shot) and a unique emotion exaggeration control that lets you dial in speech expressiveness.

Key Features

• Zero-shot voice cloning: 10-second reference audio → cloned voice
• Emotion exaggeration knob: 0.0 (flat) to 1.0+ (expressive)
• Robustness: handles long-form, awkward punctuation, and code-switching gracefully
• Streaming inference for low-latency applications
• PyTorch native, GPU and CPU support
• Pre-trained English model; community fine-tunes for other languages

Quick Start

import torchaudio
from chatterbox.tts import ChatterboxTTS

model = ChatterboxTTS.from_pretrained(device="cuda")

# Basic synthesis
wav = model.generate("Hello, this is Chatterbox speaking!")
torchaudio.save("output.wav", wav, model.sr)

# Voice cloning
wav_cloned = model.generate(
    "Voice cloning with a 10-second reference.",
    audio_prompt_path="reference.wav",
    exaggeration=0.5,
)

Install via ai-supply

npx ai-supply add chatterbox-open-source-tts

Curated mirror of the open-source Chatterbox (MIT). Get it from the source.