VieNeu-TTS

1. What is VieNeu-TTS?

VieNeu-TTS turns written text into spoken Vietnamese. You paste a paragraph, pick a voice, and get an audio file back. It was built by Phạm Nguyễn Ngọc Bảo, and unlike most online text-to-speech services, the open-source version runs on your own computer and does not need a dedicated graphics card.

The latest open-source release in the GitHub repository is VieNeu-TTS v3 Turbo, published under the Apache 2.0 license (free to use). It ships 25 preset voices covering northern, central and southern Vietnamese accents, produces 48 kHz audio, handles sentences that mix in English, and can clone a voice from a reference clip of 3 to 8 seconds.

Two different things go by the VieNeu name, so it helps to tell them apart before you install anything:

What you install	What it is	Who it suits
VieNeu Desktop app (an .exe file)	The author's Windows app, download and run	Non-technical users
The source code on GitHub	A simple demo web interface plus a library for developers	People willing to type a few commands, or anyone on macOS / Linux

According to the README, the Dubbing, Lecture, TikTok Reel and Audiobook features exist only in the official app, not in the source code. VieNeu-TTS v4, which clones voices with near-original fidelity, is proprietary and will not be open-sourced; it is offered only through vieneu.io. The README does not list prices or terms for the app or for v4, so check https://www.vieneu.io directly.

A good fit if:

• You make videos, lessons or audiobooks and need a Vietnamese narrator without recording yourself.
• You want your text processed on your own machine.
• You use Windows and want an installer you can simply double-click.

Not a good fit yet if:

• You only have a phone. An Android / iOS SDK is on the project roadmap but has not shipped.
• You use macOS and do not want to type commands. There is no macOS installer yet; the README says one is coming in a later release.

2. What do you need?

For the VieNeu Desktop app (the easy route):

• A 64-bit Windows PC. The installer is currently published for Windows x64 only.
• About 90 MB for the installer (version 0.18.9 is 89.5 MB), plus room for the app and its voices.
• An internet connection for the download.

For running from source:

• 64-bit Windows, Linux, or a Mac with Apple Silicon. These are the three platforms declared in the project's configuration file.
• uv: a tool that fetches the right Python version and all required libraries for you, so you do not have to install Python by hand. The install command is in Step 1 of Method 2. Official site: https://docs.astral.sh/uv/
• Git: the tool that downloads source code from GitHub. Get it at https://git-scm.com/downloads
• An internet connection on the first run, to download the voice model. The README does not state the total size of v3 Turbo, so leave a few GB free to be safe.
• No account and no API key.

About hardware: an ordinary CPU is enough. On the author's test machine (a 12th-generation Intel Core i5), a 10-second sentence takes about 6 seconds to generate. An NVIDIA graphics card is optional: much faster if you have one, still usable if you do not.

3. Step-by-step installation

Method 1: VieNeu Desktop app for Windows (easiest)

Step 1: Download the installer

Open either of these pages:

• The author's download page: https://www.vieneu.io/#/download
• The GitHub Releases page: https://github.com/pnnbao97/VieNeu-TTS/releases/latest

On the Releases page, scroll to "Assets" and click the file named like VieNeu_0.18.9_x64-setup.exe. Version 0.18.9 was published on 30 September 2026; by the time you read this the number may be higher, so just take the newest file.

You are on track when: your browser saves an .exe file of roughly 90 MB.

Step 2: Run the installer

Double-click the downloaded file. Windows SmartScreen may show a blue warning because the installer is not yet signed with a commercial certificate. The author's release notes say to click More info, then Run anyway.

Note: Only click "Run anyway" for a file downloaded from one of the two official pages in Step 1. Do not use installers from anywhere else.

Follow the installer screens to the end.

You are on track when: VieNeu appears in the Start menu and opens.

Method 2: Run from source (Windows, macOS, Linux)

This gives you the demo web interface of the open-source version. You will type four commands.

Step 1: Install uv

Windows (PowerShell):

powershell -c "irm https://astral.sh/uv/install.ps1 | iex"

macOS or Linux:

curl -LsSf https://astral.sh/uv/install.sh | sh

You are on track when: the command finishes without an error. Close the terminal window and open a new one so the uv command is recognized.

Step 2: Download the source code

git clone https://github.com/pnnbao97/VieNeu-TTS.git
cd VieNeu-TTS

You are on track when: the download progress finishes and your prompt is now inside the VieNeu-TTS folder.

Step 3: Install the required libraries

For CPU-only machines and Macs:

uv sync

For machines with an NVIDIA card, if you want faster generation:

uv sync --extra cuda

Run only one of the two. If you are not sure whether you have an NVIDIA card, use the first one.

You are on track when: uv finishes downloading and returns you to the prompt with no red error lines.

Step 4: Start the web interface

uv run vieneu-web

You are on track when: the terminal prints the address http://127.0.0.1:7860. Open it in your browser. Keep the terminal window open while you work; closing it shuts the interface down.

Method 3: For people who write Python

If you want to call VieNeu from your own program, the library installs with one command:

pip install vieneu

Code samples and the extra steps for NVIDIA cards are in the project README: https://github.com/pnnbao97/VieNeu-TTS

4. First use

With the web interface (Method 2):

1. Open http://127.0.0.1:7860 in your browser.
2. In the model selector, leave VieNeu-TTS-v3-Turbo chosen.
3. Pick a voice. The default is Hải Đăng. The authors suggest starting with these ten: Adam bựa, Trúc Ly, Thiện Minh, Mai Anh, Hải Đăng, Thùy Dung, Thiền Tâm Đức, Ngọc Huyền, Quang Sơn, Ngọc Trân.
4. Paste a short Vietnamese sentence, for example "Xin chào, đây là giọng đọc chạy trên máy của mình."
5. Click the generate button and wait.

The first generation takes noticeably longer because the model has to be downloaded and then loaded into memory (loading alone took about 19 seconds on the author's test machine). Later runs are quicker.

The result is an audio clip you can play right on the page.

Two more things to try:

• Emotion cues (experimental): type [cười] (chuckle), [thở dài] (sigh) or [hắng giọng] (clearing the throat) straight into the sentence.
• Voice cloning: supply a reference recording of 3 to 8 seconds. The tool removes background noise and trims the clip to at most 8 seconds on its own.

Note: Only clone your own voice or the voice of someone who has agreed. The author keeps v4 closed precisely because of how easily this technology can be misused.

With the VieNeu Desktop app: open it from the Start menu, paste your text, pick a voice and generate. The exact layout changes between versions, so follow the prompts inside the app.

5. Common problems and fixes

Symptom	Cause	Fix
Windows shows "Windows protected your PC" when you run the installer	The installer is not signed with a commercial certificate yet	Click More info, then Run anyway (only for a file from an official page)
Typing uv gives "not recognized" or "command not found"	The terminal has not picked up the newly installed command	Close the window, open a new one, and continue from where you stopped
Typing git gives "not recognized" or "command not found"	Git is not installed	Install Git from https://git-scm.com/downloads and reopen the terminal
The first generation seems stuck for a long time	The model is still downloading	Wait for it to finish and stay online. It will not be downloaded again
The browser cannot open http://127.0.0.1:7860	The interface has not finished starting, or the terminal window was closed	Run uv run vieneu-web again and wait for the address to appear before opening the browser
Audio is garbled after switching to int8 mode	That mode needs a CPU with VNNI instructions; older CPUs lack them	Go back to the default fp32 mode
Generation is too slow on a weak machine	v3 Turbo is heavy for that hardware	Developers can try v3 Nano: about 3 times faster but lower quality, 24 kHz audio, 11 voices
Changing the style setting has no effect	style is deprecated on v3 Turbo	Change the reading style by picking another voice or another reference clip

If your problem is not listed, search or ask in the project's Issues: https://github.com/pnnbao97/VieNeu-TTS/issues

6. Uninstalling / updating

VieNeu Desktop app:

• Update: the running app tells you when a new version is available. You can also download the newer installer from the Releases page and install it over the current one.
• Uninstall: open Windows Settings, go to Apps, then Installed apps, find VieNeu and choose Uninstall.

Source install:

To update, run these inside the VieNeu-TTS folder:

git pull
uv sync

To remove it, close the terminal running the interface and delete the whole VieNeu-TTS folder. Downloaded models live in the Hugging Face cache (by default the .cache/huggingface folder inside your user folder); delete that too if you want the disk space back.

Python library:

pip install -U vieneu

pip uninstall vieneu

7. Frequently asked questions

Does it cost money? The open-source v3 Turbo in the GitHub repository is Apache 2.0 licensed, and the README describes it as "Free to use". Version 4 and the app-only features are the author's own product; see https://www.vieneu.io for prices and terms.

Do I need an internet connection? Yes for installing and for the first run, when the model is downloaded. The open-source version is designed to generate speech on the device itself.

Is my text or voice sent anywhere? With the open-source v3 Turbo, generation happens on your computer. Version 4 is different: it is delivered through vieneu.io and the VieNeu API, which means it runs on the author's servers. Read the terms on that site before using it.

Will it run without a dedicated graphics card? Yes. On a 12th-generation Core i5 CPU it generates speech roughly twice as fast as the audio plays. Weaker machines will be slower.

Can I use it on a phone? Not yet. An Android and iOS SDK is on the roadmap but is not marked as done.