Skip to content



We assume no responsibility for any illegal use of the codebase. Please refer to the local laws regarding DMCA (Digital Millennium Copyright Act) and other relevant laws in your area.

This codebase is released under the BSD-3-Clause license, and all models are released under the CC-BY-NC-SA-4.0 license.


  • GPU Memory: 4GB (for inference), 16GB (for fine-tuning)
  • System: Linux, Windows

Windows Setup

Windows professional users may consider WSL2 or Docker to run the codebase.

Non-professional Windows users can consider the following methods to run the codebase without a Linux environment (with model compilation capabilities aka torch.compile):

  1. Unzip the project package.
  2. Click install_env.bat to install the environment.
    • You can decide whether to use a mirror site for downloads by editing the USE_MIRROR item in install_env.bat.
    • USE_MIRROR=false downloads the latest stable version of torch from the original site. USE_MIRROR=true downloads the latest version of torch from a mirror site. The default is true.
    • You can decide whether to enable the compiled environment download by editing the INSTALL_TYPE item in install_env.bat.
    • INSTALL_TYPE=preview downloads the preview version with the compiled environment. INSTALL_TYPE=stable downloads the stable version without the compiled environment.
  3. If step 2 has USE_MIRROR=preview, execute this step (optional, for activating the compiled model environment):
    1. Download the LLVM compiler using the following links:
    2. Download and install the Microsoft Visual C++ Redistributable package to resolve potential .dll missing issues.
    3. Download and install Visual Studio Community Edition to obtain MSVC++ build tools, resolving LLVM header file dependencies.
      • Visual Studio Download
      • After installing Visual Studio Installer, download Visual Studio Community 2022.
      • Click the Modify button as shown below, find the Desktop development with C++ option, and check it for download.
  4. Double-click start.bat to enter the Fish-Speech training inference configuration WebUI page.
    • (Optional) Want to go directly to the inference page? Edit the API_FLAGS.txt in the project root directory and modify the first three lines as follows:
      # --api
      # --listen ...
    • (Optional) Want to start the API server? Edit the API_FLAGS.txt in the project root directory and modify the first three lines as follows:
      # --infer
      --listen ...
  5. (Optional) Double-click run_cmd.bat to enter the conda/python command line environment of this project.

Linux Setup

# Create a python 3.10 virtual environment, you can also use virtualenv
conda create -n fish-speech python=3.10
conda activate fish-speech

# Install pytorch
pip3 install torch torchvision torchaudio

# Install fish-speech
pip3 install -e .

# (Ubuntu / Debian User) Install sox
apt install libsox-dev


  • 2024/05/10: Updated Fish-Speech to 1.1 version, implement VITS decoder to reduce WER and improve timbre similarity.
  • 2024/04/22: Finished Fish-Speech 1.0 version, significantly modified VQGAN and LLAMA models.
  • 2023/12/28: Added lora fine-tuning support.
  • 2023/12/27: Add gradient checkpointing, causual sampling, and flash-attn support.
  • 2023/12/19: Updated webui and HTTP API.
  • 2023/12/18: Updated fine-tuning documentation and related examples.
  • 2023/12/17: Updated text2semantic model, supporting phoneme-free mode.
  • 2023/12/13: Beta version released, includes VQGAN model and a language model based on LLAMA (phoneme support only).