Fun-CosyVoice3-0.5B

Fun-CosyVoice3-0.5B

Turns written text into speech in nine languages and many Chinese dialects, with zero-shot voice cloning, in a Gradio demo for anyone testing speech synthesis.

Made by FunAudioLLM on Hugging Face. These notes are Mindket's, written from its Space page and README; the app's own files and license are on Hugging Face.

About

What it is: Fun-CosyVoice3-0.5B is a text-to-speech system built on large language models, from the FunAudioLLM team. This Space is a Gradio demo of it that runs on GPU hardware at Hugging Face. It is meant for speech synthesis across languages, and it can imitate a voice from a short sample without extra training.

Who it suits: People who need natural synthetic speech in several languages, and people who want to hear how a recent open speech model sounds.

Industry | Who uses it | How they use it | Example
Media | Video and audio producers | Add narration in different languages | Read a script aloud in Spanish
Software | Developers of voice apps | Test a speech model before adding it to a product | Listen to streaming output for latency
Education | Language teachers | Hear a text read in a given language or dialect | Play a Sichuan dialect version of a sentence
Research | Speech researchers | Compare synthesis quality and pronunciation control | Try Pinyin or phoneme hints on a hard word

People who only need a plain screen reader can skip it, since this is a model demo.

Where to use it:

• A browser, by visiting the Space on Hugging Face
• Windows, macOS or Linux with Python, for a copy on your own computer
• A local Gradio page, once app.py is running

Use case | Who | What they get back
Read text aloud | Producers and teachers | Speech in one of nine languages or a Chinese dialect
Clone a voice from a sample | Developers | Speech that sounds like the supplied voice
Control delivery | Researchers | Speech shaped by emotion, speed or volume instructions

How to set it up:

• Python and pip on Windows, macOS or Linux
• The Python packages in requirements.txt
• Gradio 6.1.0, the version the Space uses
• A GPU, since the Space is set to zero-a10g

1. Install the Hugging Face hub package.
2. Download the Space with the hf command.
3. Move into the new folder.
4. Install gradio at the version the Space uses, then the listed packages.
5. Start the app with Python.

pip install -U huggingface_hub
hf download FunAudioLLM/Fun-CosyVoice3-0.5B --repo-type space --local-dir Fun-CosyVoice3-0.5B
cd Fun-CosyVoice3-0.5B
pip install gradio==6.1.0
pip install -r requirements.txt
python app.py

How to use it: Online, visit the Space on Hugging Face, enter the text to speak and add a voice sample if the page asks for one. On your own computer, run the start command and open the local page Gradio serves.

python app.py

Tool | What it does
app.py | Starts the Gradio interface
requirements.txt | Lists the Python packages to install
zero_shot_prompt.wav | A sample voice file kept in the repository
cosyvoice | The model code folder

Example: You type a sentence in English and supply a short voice sample. The demo reads the sentence aloud, aiming to sound like the sample voice.

For newcomers and experts: If you are new, try a short sentence in a language you know and listen to the result. Zero-shot means the voice is copied from a sample without training a new model on it.

If you know the field, the README lists 9 languages and more than 18 Chinese dialects and accents. It also lists pronunciation inpainting with Chinese Pinyin and English CMU phonemes, text normalization without a classic frontend, and bi-streaming with latency as low as 150ms. Instructions can set language, dialect, emotion, speed and volume.

You enter text and a voice sample
The language model plans the speech
The speech model renders the audio
You play the result in the browser

Good to know:

Title: Character error rate on the Chinese test set, lower is better
MiniMax-Speech: 0.83%
Fun-CosyVoice3-0.5B-2512_RL: 0.81%
Fun-CosyVoice3-0.5B-2512: 1.21%
Seed-TTS: 1.12%
CosyVoice2: 1.45%
F5-TTS: 1.53%

• Languages: nine common languages plus 18 or more Chinese dialects and accents
• Hardware: the Space is set to zero-a10g
• License: apache-2
• Cost to run: GPU hardware, no API key or paid service is named
• Data it reaches: the text and voice samples you give it, plus the models it fetches from Hugging Face

On Hugging Face, 13 November 2026

• Likes: 87
• Built with: gradio
• License: apache-2.0

Links

• Space: https://huggingface.co/spaces/FunAudioLLM/Fun-CosyVoice3-0.5B
• Files: https://huggingface.co/spaces/FunAudioLLM/Fun-CosyVoice3-0.5B/tree/main

Try it

Open the Space in your browser, or copy it to your own Hugging Face account from its page. Nothing in this kit runs by itself.