VoxCPM — Tokenizer-Free Multilingual Voice Cloning TTS
OpenBMB's VoxCPM2 model, embedded from the official Hugging Face Space
Try VoxCPM online for multilingual text-to-speech, voice design, controllable voice cloning and 48 kHz AI speech. This page rewrites public VoxCPM2 facts into an independent, crawlable guide with the official demo embedded below.
This is the official VoxCPM-Demo Space hosted on Hugging Face by OpenBMB. The first run may be slower while the hosted demo wakes up. Avoid confidential text or private voice samples in any public Space.
What is VoxCPM?
VoxCPM is an open text-to-speech model family from OpenBMB. The current VoxCPM2 release focuses on natural multilingual speech, creative voice design and true-to-life voice cloning, while avoiding a discrete speech-tokenizer pipeline.
According to the public VoxCPM2 model card and project documentation, VoxCPM2 is a 2B-parameter tokenizer-free diffusion autoregressive TTS model trained on more than 2 million hours of multilingual speech. It supports 30 languages, accepts short reference audio for cloning, and outputs 48 kHz speech through AudioVAE V2.
The embedded Space is useful for quick evaluation because it exposes the workflows teams actually test first: direct multilingual TTS, natural-language voice design, controllable cloning with style guidance, and higher-fidelity cloning with reference audio plus transcript.
Why VoxCPM2 stands out
VoxCPM combines multilingual generation, voice design, cloning and production-oriented serving in one open model workflow.
Tokenizer-free diffusion autoregressive TTS
VoxCPM2 directly generates continuous speech representations through a diffusion autoregressive architecture instead of relying on discrete speech tokens.
30-language multilingual speech
The model card lists 30 supported languages, including Chinese, English, Japanese, Korean, Spanish, French, German, Portuguese, Hindi, Arabic, Vietnamese and more.
Voice design from natural language
Describe a speaker with attributes such as gender, age, tone, emotion or pace, then let VoxCPM2 create a new synthetic voice without reference audio.
Controllable and ultimate voice cloning
Upload a short reference clip to preserve timbre, add style guidance to steer expression, or provide a transcript for higher-fidelity continuation-style cloning.
48 kHz output and streaming path
VoxCPM2 accepts 16 kHz reference audio and outputs 48 kHz speech; the docs report real-time streaming with standard inference and faster Nano-vLLM serving.
Open and safety-aware release
The public release is Apache-2.0, but the project explicitly forbids impersonation, fraud and disinformation. Label AI-generated audio clearly.
Try VoxCPM in four steps
The hosted Space gives you the core VoxCPM2 workflows without a local setup.
Open the VoxCPM Space
Use the embedded Hugging Face Space above or open it in a new tab. Cold starts can take extra time while the hosted environment loads dependencies and the model.
Enter text in a supported language
Paste your script directly. VoxCPM2 is designed for multilingual input without a separate language tag, so start with a short sentence in your target language.
Choose voice design or cloning
For voice design, put a natural-language speaker description before the text. For cloning, upload a clean reference clip and add style guidance only when you want to change pace or emotion.
Tune generation and listen
Adjust inference steps, guidance, denoise or style settings in the demo, generate a short sample, then iterate until pronunciation, pacing and speaker style are right.
VoxCPM capabilities at a glance
Key public facts from the VoxCPM2 Hugging Face model card, Space metadata, documentation and project demo page, rewritten for quick evaluation.
What you can build with VoxCPM
VoxCPM is most useful when a voice project needs multilingual breadth, controllable expression and the option to self-host.
Multilingual product narration
Create draft voiceovers for product tours, lessons and explainers across many languages before recording final human narration.
Voice-agent prototyping
Explore voice personas, speed, emotional tone and 48 kHz output quality before connecting a production voice agent stack.
Dubbing and localization tests
Evaluate pronunciation, rhythm and expressiveness for localized scripts in languages that are often underserved by older TTS systems.
Consent-based voice cloning
Clone a speaker only when you have permission, then use controllable guidance to test different emotions or pacing while preserving timbre.
Synthetic character exploration
Use voice design when you need a new fictional or brand-safe voice without cloning a real person.
Developer evaluation
Compare VoxCPM against other open TTS systems, test the Python package, or study the Nano-vLLM serving path for lower-latency experiments.
Tips for better VoxCPM results
- Start with short text so you can quickly evaluate pronunciation, rhythm and voice consistency before generating long passages.
- Use clean, single-speaker reference audio for cloning; noisy clips or multiple speakers make speaker similarity harder to judge.
- If a designed voice misses the target, rewrite the speaker description with concrete attributes such as age, pace, emotion and vocal texture.
- For high-fidelity cloning, provide the reference transcript when the demo asks for it, because transcript-aligned context helps preserve the original delivery.
- Avoid confidential scripts or private voice samples in a public hosted Space; self-host VoxCPM2 for sensitive production work.
- Do not use VoxCPM for impersonation, fraud or unlabeled synthetic media. Obtain consent before cloning any real voice.
VoxCPM frequently asked questions
Short answers for people evaluating VoxCPM for multilingual TTS, voice cloning, voice design and self-hosted AI voice workflows.
What is VoxCPM?
VoxCPM is OpenBMB's tokenizer-free text-to-speech model family. VoxCPM2 is the current 2B-parameter release for multilingual speech generation, voice design and voice cloning.
Can I try VoxCPM online?
Yes. This page embeds the official VoxCPM-Demo Hugging Face Space, so you can test VoxCPM2 in the browser without installing the Python package first.
What languages does VoxCPM2 support?
The public VoxCPM2 model card lists 30 languages, including Chinese, English, Japanese, Korean, Spanish, French, German, Portuguese, Hindi, Arabic, Vietnamese and others, plus several Chinese dialects.
Does VoxCPM support voice cloning?
Yes. VoxCPM2 supports controllable voice cloning from reference audio and an ultimate cloning workflow that can use both the reference clip and its transcript for higher fidelity. Clone only voices you own or have permission to use.
What is VoxCPM voice design?
Voice design lets you describe a desired speaker in natural language — for example age, gender, tone, emotion or pace — and generate speech in that newly designed voice without uploading a reference recording.
Is the embedded VoxCPM demo private?
No. The embedded app is a public Hugging Face Space run outside Whisper AI. Do not submit confidential text or private voice samples there; self-host the model for sensitive work.
Try VoxCPM in your browser
Generate multilingual speech, test voice design, and evaluate consent-based cloning in the official Hugging Face demo before installing VoxCPM locally.