Spark-TTS is a recent open-source voice cloning project that has gained significant attention, jointly developed by several universities including the Hong Kong University of Science and Technology, Northwestern Polytechnical University, and Shanghai Jiao Tong University. After local testing, its performance is comparable to F5-TTS.

Spark-TTS supports Chinese and English voice cloning, and the installation and deployment process is not complicated. This article will detail how to install and deploy it, and modify it to be compatible with the F5-TTS API interface, so that it can be directly used in the F5-TTS dubbing channel of the pyVideoTrans software.

Prerequisites: Ensure that Python version 3.10, 3.11, or 3.12 is installed.

If you haven't installed it yet, please refer to the previous article for installation instructions. We won't repeat them here.

1. Download Spark-TTS Source Code

First, create a folder composed of English letters or numbers on a non-system disk, such as D:/spark. The reason for requiring English letters, numbers, and a non-system disk is to avoid potential errors related to Chinese characters, permissions, etc. as much as possible.

Then, visit the official Spark-TTS code repository: https://github.com/SparkAudio/Spark-TTS

As shown in the figure below, click to download the ZIP package of the source code:

After downloading and extracting, copy all files and folders to the D:/spark folder. The directory structure after copying should be as shown in the following figure:

2. Create a Virtual Environment and Install Dependencies

• Create a Virtual Environment

Enter cmd in the address bar of the folder and press Enter. In the black terminal window that pops up, execute the following command:

python -m venv venv

As shown in the figure:

After execution, a venv folder will be added to the D:/spark directory:

Note: If you are prompted with python is not an internal or external command during execution, it means that Python is not installed or has not been added to the system environment variables. Please refer to relevant articles to install Python.

Next, execute venv\scripts\activate to activate the virtual environment. After activation, (venv) will appear at the beginning of the terminal line, indicating successful activation. All subsequent commands need to be executed in this environment. Please check whether it has been activated before each execution.

• Install Dependencies

In the activated virtual environment, continue to execute the following command in the terminal to install all dependencies:

pip install -r requirements.txt

The installation process may take a long time, please be patient.

3. Download Models

The models required for open-source AI projects are usually hosted on Hugging Face (huggingface.co). Since this website is blocked in China, you need to use a VPN to download the models. Please make sure that the VPN environment is configured and the system proxy is set.

Create a text file named down.txt in the current directory D:/spark, copy and paste the following code into the file and save it:

from huggingface_hub import snapshot_download
snapshot_download("SparkAudio/Spark-TTS-0.5B", local_dir="pretrained_models/Spark-TTS-0.5B")
print('下载完成')

Then, execute the following command in the terminal window with the virtual environment activated:

python down.txt

Check if there is a (venv) prefix on the command line:

Wait for the terminal to indicate that the download is complete.

If output similar to the following information, it means that the network connection is incorrect, and the VPN environment may not be configured correctly:

Returning existing local_dir `pretrained_models\Spark-TTS-0.5B` as remote repo cannot be accessed in `snapshot_download` ((MaxRetryError("HTTPSConnectionPool(host='huggingface.co', port=443): Max retries exceeded with url: /api/models/SparkAudio/Spark-TTS-0.5B/revision/main (Caused by ConnectTimeoutError(<urllib3.connection.HTTPSConnection object at 0x000001BC4C8A4430>, 'Connection to huggingface.co timed out. (connect timeout=None)'))"), '(Request ID: aa61d1fb-ffc7-4479-9a99-2258c1bc0aee)')).

4. Start the Web Interface

After the model is downloaded, you can start and open the Web interface.

Execute the following command in the terminal with the virtual environment activated:

python webui.py

Wait for the following information to appear, indicating that the startup is complete:

At this point, you can open the address http://127.0.0.1:7860 in your browser. The Web interface is shown in the following figure:

5. Voice Cloning Test

As shown in the figure below, select an audio file with the voice you want to clone (duration 3-10 seconds, clear pronunciation, clean background).

Then, enter the text corresponding to the audio in the Text of prompt speech on the right. Enter the text you want to generate on the left, and finally click the Generate button at the bottom to start execution.

After execution, it is shown in the figure below.

6. Use in pyVideotrans Software

Spark-TTS is very similar to F5-TTS. With a simple modification, Spark-TTS can be directly used in the F5-TTS dubbing channel of pyVideotrans.

• Open the webui.py file and paste the following code above line 135:

    def basic_tts(gen_text_input, ref_text_input, ref_audio_input,remove_silence=None,speed_slider=None):
        """
        Gradio callback to clone voice using text and optional prompt speech.
        - text: The input text to be synthesised.
        - prompt_text: Additional textual info for the prompt (optional).
        - prompt_wav_upload/prompt_wav_record: Audio files used as reference.
        """
        prompt_speech = ref_audio_input
        prompt_text_clean = None if len(ref_text_input) < 2 else ref_text_input

        audio_output_path = run_tts(
            gen_text_input,
            model,
            prompt_text=prompt_text_clean,
            prompt_speech=prompt_speech
        )
        return audio_output_path,prompt_text_clean

Special Note: Python code uses spaces for indentation and alignment, otherwise the code will report an error. To avoid errors, it is recommended not to open the webui.py file with Notepad, but to use a professional code editor, such as Notepad++ or VSCode.

• Then, find the generate_buttom_clone = gr.Button("Generate") code around line 190. Paste the following code above it, and pay attention to alignment:

generate_buttom_clone2 = gr.Button("Generate2",visible=False)
generate_buttom_clone2.click(
       basic_tts,
       inputs=[
          text_input,
          prompt_text_input,
          prompt_wav_upload,
          text_input,
          text_input
       ],
       outputs=[audio_output,prompt_text_input],
       api_name="basic_tts"
 )

• After saving the file, restart webui.py:

python webui.py

• Fill in the address http://127.0.0.1:7860 into the API address of "Menu" -> "TTS Settings" -> "F5-TTS" in the pyVideotrans software, and you can start using it. The position and filling method of the reference audio are the same as the usage method of F5-TTS.