VoxCPM: Tokenizer-Free TTS for Multilingual Speech, Voice Design, and Cloning

Introduction

VoxCPM2 is a cutting-edge, tokenizer-free Text-to-Speech (TTS) system developed by OpenBMB. It directly generates continuous speech representations using an end-to-end diffusion autoregressive architecture, bypassing discrete tokenization for highly natural and expressive synthesis. The latest major release, VoxCPM2, is a 2B parameter model trained on over 2 million hours of multilingual speech data, now supporting 30 languages. It introduces innovative features like Voice Design, Controllable Voice Cloning, and delivers 48kHz studio-quality audio output. Built on a MiniCPM-4 backbone, VoxCPM2 is fully open-source and commercial-ready under the Apache-2.0 license.

Installation

Getting started with VoxCPM is straightforward. Ensure you have Python ? 3.10 (<3.13), PyTorch ? 2.5.0, and CUDA ? 12.0 installed.

pip install voxcpm

For more detailed installation instructions, refer to the official documentation.

Examples

VoxCPM provides a flexible Python API for various speech synthesis tasks.

Text-to-Speech

Generate speech from text with ease:

from voxcpm import VoxCPM
import soundfile as sf

model = VoxCPM.from_pretrained("openbmb/VoxCPM2", load_denoiser=False)

wav = model.generate(
    text="VoxCPM2 is a powerful multilingual speech synthesis system.",
    cfg_value=2.0,
    inference_timesteps=10,
)
sf.write("demo.wav", wav, model.tts_model.sample_rate)
print("saved: demo.wav")

Voice Design

Create a brand-new voice from a natural-language description, no reference audio needed. Simply put the description in parentheses at the start of your text:

wav = model.generate(
    text="(A young woman, gentle and sweet voice)Hello, welcome to VoxCPM2!",
    cfg_value=2.0,
    inference_timesteps=10,
)
sf.write("voice_design.wav", wav, model.tts_model.sample_rate)

Controllable Voice Cloning

Clone any voice from a short reference clip, with optional style guidance to steer emotion, pace, and expression while preserving the original timbre:

wav = model.generate(
    text="(slightly faster, cheerful tone)This is a cloned voice with style control.",
    reference_wav_path="path/to/voice.wav",
    cfg_value=2.0,
    inference_timesteps=10,
)
sf.write("controllable_clone.wav", wav, model.tts_model.sample_rate)

Why Use VoxCPM?

VoxCPM2 stands out with several key advantages for speech generation:

• 30-Language Multilingual Support: Synthesize text in any of the 30 supported languages without needing language tags.
• Creative Voice Design: Generate unique voices purely from natural-language descriptions, eliminating the need for reference audio.
• Controllable Voice Cloning: Clone voices from short clips and fine-tune style, emotion, and pace while maintaining the original timbre.
• 48kHz High-Quality Audio: Output studio-quality audio directly, with built-in super-resolution.
• Real-Time Streaming: Achieve low Real-Time Factor (RTF) for efficient, high-throughput serving, especially with Nano-vLLM or vLLM-Omni.
• Fully Open-Source & Commercial-Ready: Released under the Apache-2.0 license, making it free for commercial use.

Links

Explore VoxCPM further through these official resources:

• GitHub Repository
• Live Playground Demo
• Official Documentation
• VoxCPM2 Model on Hugging Face