Confucius4-TTS Voice Clone

Confucius4-TTS Voice Clone

Clones a voice from a short reference clip and speaks your text in it, with automatic language detection and optional transcription of the clip. It suits people testing text-to-speech and voice cloning.

Made by CharlesW 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: This Space lets you clone a voice with Confucius4-TTS, using a page split into two columns. You enter the text to speak and give a reference voice, and it generates speech in that voice. The interface is in English, and the language of your text is detected automatically. It runs on zero-a10g hardware at Hugging Face.

Who it suits: It suits people who want to test voice cloning from a short clip, and developers who want to see how reference transcription and language routing are handled.

Industry | Who uses it | How they use it | Example
Content creation | Video and podcast makers | Voice a script in a chosen reference voice | Read a short line in the voice of a clean sample
AI research | Speech researchers | Compare audio-only and transcript-guided cloning | Run the same clip with and without a transcript
Software | Developers | Study a Gradio TTS app with automatic recognition | Read app.py and the routing files

You can skip it if you need a production service, since the README calls the transcript path experimental.

Where to use it: You can use it in these places:

• A browser: open the Space's page on Hugging Face and nothing needs installing.
• Windows, macOS or Linux: run your own copy with Python 3.10 and a GPU.

Use case | Who | What they get back
Voice cloning | Creators | Speech in your text spoken in the reference voice
Cross-language use | Researchers | Speech from a reference in one language and text in another
Transcript-guided cloning | Developers | Generation that uses the clip together with its transcript

How to set it up: You need these first for a local copy:

• Python 3.10 and pip
• The system packages listed in packages.txt, which Hugging Face installs with apt-get
• A GPU, since the model and a semantic codec use GPU memory

The online page needs none of this. For your own copy, follow the steps below.

1. Install the Hugging Face tools.
2. Download the Space's files into a folder.
3. Move into that folder.
4. Install Gradio at the version the Space uses, then the other packages.
5. Start the app.

pip install -U huggingface_hub
hf download CharlesW/Confucius4-TTS --repo-type space --local-dir Confucius4-TTS
cd Confucius4-TTS
pip install gradio==5.50.0
pip install -r requirements.txt
python app.py

How to use it: Open the Space in a browser, or start your own copy with the command above. Type the text to speak, then upload or record a reference voice. Clear speech of 3 to 10 seconds is recommended, and the final clip must be 1 to 15 seconds. Click Generate when the transcript step is done. The transcript is optional.

Tool | What it does
Target text | Takes the text to speak and detects its language, which you can override
Reference audio | Takes an uploaded or recorded voice and starts transcription on its own
Reference transcript | Optional text of the clip that you can correct or clear
Scissors tool | Lets you pick a segment by hand from a long upload
Audio Only | Ignores the transcript and uses the audio alone
Generate | Creates the speech

Example: You type a sentence, then record a few seconds of clear speech as the reference. The app transcribes it, you correct any slip, and you click Generate. It returns your sentence spoken in the reference voice.

For newcomers and experts: If you are new to this, try Audio Only first with a short, clear clip. Add the transcript later if you want to see the difference.

If you know the field, ordinary speech generation goes through the generate call of the published ConfuciusTTS package. The transcript path adds a semantic codec from MaskGCT and acoustic continuation, and the README calls it experimental with no confirmed quality on every recording. Transcription uses Whisper large-v3-turbo on CPU, and clips over 30 seconds get timestamp-aware transcription.

You enter the text to speak
You add a reference voice
Whisper transcribes the clip
The model clones the voice
You get the audio

Good to know: Recognition weights download on the first audio upload, and a recognition failure does not stop audio-only generation. Chinese transcripts appear in simplified characters, and you may still need to fix mistakes.

• License: apache-2 on the Space page, though the model repository is CC-BY-NC-4.0 and the Space's own license does not change that
• Cost to run: none named
• Data it reaches: the reference audio you upload or record, and model files downloaded from Hugging Face
• Not for: uses outside the CC-BY-NC-4.0 terms of the model

On Hugging Face, 13 November 2026

• Likes: 10
• Built with: gradio
• License: apache-2.0

Links

• Space: https://huggingface.co/spaces/CharlesW/Confucius4-TTS
• Files: https://huggingface.co/spaces/CharlesW/Confucius4-TTS/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.