目录

• 前言
• 一、安装 Conda
  • Download Spark-TTS
• 二、创建 Conda 环境
• 三、安装依赖
• 四、安装 PyTorch
• 下载 Spark-TTS 模型
• 六、运行 Spark-TTS
  • Web UI (Recommended)
• 七、调试与问题
• 八、启动 webui.py 失败可能原因
  • 1. 查看当前安装的 torch 和 torchvision 版本
  • 2. 检查版本是否匹配
  • 3. 调整版本以确保匹配
    • 方法 1：升级或降级 torchvision
    • 方法 2：升级或降级 torch
    • 方法 3：使用 Conda 管理版本
  • 4. 验证安装
• 九、CUDA 运行版本对不上

前言

5月时只是小朋友为了智能体会话可以用周深的声音对话，所以折腾了台机器，搭建了一个本地的 Spark-TTS。折腾了这老版本 CUDA、PyTorch 的匹配，多次下载安装不同版本和驱动，抽几个晚上最终定格了版本匹配。

一、安装 Conda

如果还没有安装 Conda 的话。

• Download Miniconda and install it.
• Make sure to check “Add Conda to PATH” during installation.

Download Spark-TTS

You have two options to get the files:

Option 1 (Recommended for Windows): Download ZIP manually

• Go to Spark-TTS GitHub
• Click “Code” > “Download ZIP”, then extract it.

Option 2: Use Git (Optional)

• If you prefer using Git, install Git and run:

  git clone https://github.com/SparkAudio/Spark-TTS.git

二、创建 Conda 环境

Open Command Prompt (cmd) and run:

conda create -n sparktts python=3.12 -y conda activate sparktts

This creates and activates a Python 3.12 environment for Spark-TTS.

三、安装依赖

Inside the Spark-TTS folder (whether from ZIP or Git), run:

pip install -r requirements.txt

四、安装 PyTorch

注：自动检测 CUDA 或 CPU，会自动安装对应版本的 PyTorch。

pip install torch torchvision torchaudio --index-url https://pytorch.org/get-started/previous-versions/

# OR Manually install a specific CUDA version (if needed)
# pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121
# pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118  # Older GPUs

可能有些电脑比较旧，安装的 cuda 版本比较低，导致没有找到正确的 torch 版本。

(sparktts) I:\AI-src\Spark-TTS>pip install torch torchvision torchaudio --index-url https://pytorch.org/get-started/previous-versions/
Looking in indexes: https://pytorch.org/get-started/previous-versions/
Requirement already satisfied: torch in d:\anaconda3\envs\sparktts\lib\site-packages (2.5.1)
ERROR: Could not find a version that satisfies the requirement torchvision (from versions: none)
ERROR: No matching distribution found for torchvision

解决：

先检查本地 cuda 的版本：

(sparktts) I:\AI-src\Spark-TTS>nvcc --version
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2022 NVIDIA Corporation
Built on Wed_Sep_21_10:41:10_Pacific_Daylight_Time_2022
Cuda compilation tools, release 11.8, V11.8.89
Build cuda_11.8.r11.8/compiler.31833905_0

注：CUDA 一开始 runtime 版本是 10，后来可支持最新升级到 12.1，但由于 CUDA 12.1 一直安装的+cpu 版本，不能支持 CUDA，所以只能使用 11.8 CUDA 版本。

根据版本 11.8 指定 index-url 重新安装：

pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118

确认

(sparktts) I:\AI-src\Spark-TTS>python ver.py
PyTorch version: 2.6.0+cu118
TorchVision version: 0.21.0+cu118

下载 Spark-TTS 模型

There are two ways to get the model files. Pick one:

Option 1 (Recommended): Using Python
Create a new file in the Spark-TTS folder called download_model.py, paste this inside, and run it:

from huggingface_hub import snapshot_download import os

# Set download path model_dir = “pretrained_models/Spark-TTS-0.5B”

# Check if model already exists if os.path.exists(model_dir) and len(os.listdir(model_dir)) > 0: print(“Model files already exist. Skipping download.") else: print(“Downloading model files…") snapshot_download( repo_id="SparkAudio/Spark-TTS-0.5B”, local_dir=model_dir, resume_download=True # Resumes partial downloads ) print(“Download complete!")

Run it with:

python download_model.py

✅ Option 2: Using Git (If You Installed It)

mkdir pretrained_models git clone https://huggingface.co/SparkAudio/Spark-TTS-0.5B pretrained_models/Spark-TTS-0.5B

Either method works—choose whichever is easier for you.

建议直接使用 git：

(sparktts) I:\AI-src\Spark-TTS\pretrained_models>git clone https://huggingface.co/SparkAudio/Spark-TTS-0.5B pretrained_models/Spark-TTS-0.5B
Cloning into 'pretrained_models/Spark-TTS-0.5B'...
remote: Enumerating objects: 80, done.
remote: Counting objects: 100% (76/76), done.
remote: Compressing objects: 100% (76/76), done.
remote: Total 80 (delta 21), reused 0 (delta 0), pack-reused 4 (from 1)
Unpacking objects: 100% (80/80), 3.63 MiB | 1.43 MiB/s, done.
Updating files: 100% (31/31), done.
Filtering content: 100% (4/4), 3.66 GiB | 8.44 MiB/s, done.

六、运行 Spark-TTS

Web UI (Recommended)

For an interactive browser-based interface, run:

python webui.py

This launches a local web server where you can enter text and generate speech or clone a voice.

(sparktts) I:\AI-src\Spark-TTS>python webui.py
D:\anaconda3\envs\sparktts\Lib\site-packages\torch\nn\utils\weight_norm.py:143: FutureWarning: `torch.nn.utils.weight_norm` is deprecated in favor of `torch.nn.utils.parametrizations.weight_norm`.
  WeightNorm.apply(module, name, dim)
Missing tensor: mel_transformer.spectrogram.window
Missing tensor: mel_transformer.mel_scale.fb
* Running on local URL:  http://0.0.0.0:7860

To create a public link, set `share=True` in `launch()`.
Setting `pad_token_id` to `eos_token_id`:None for open-end generation.

七、调试与问题

🔎 Before Asking for Help
Many common issues are already covered in existing discussions, documentation, or online resources. Please:

• Search GitHub issues first 🕵️‍♂️
• Check the documentation 📖
• Google or use AI tools (ChatGPT, DeepSeek, etc.)

If you still need help, please explain what you’ve already tried so we can assist you better!

Now you’re good to go! 🚀🔥

Happy TTS-ing.

八、启动 webui.py 失败可能原因

torch 和 torchvision 版本不匹配

错误信息如下：

(sparktts) I:\AI-src\Spark-TTS>python webui.py
Traceback (most recent call last):
  File "D:\anaconda3\envs\sparktts\Lib\site-packages\transformers\utils\import_utils.py", line 1778, in _get_module
    return importlib.import_module("." + module_name, self.__name__)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\anaconda3\envs\sparktts\Lib\importlib\__init__.py", line 90, in import_module
    return _bootstrap._gcd_import(name[level:], package, level)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "<frozen importlib._bootstrap>", line 1387, in _gcd_import
  File "<frozen importlib._bootstrap>", line 1360, in _find_and_load
  File "<frozen importlib._bootstrap>", line 1331, in _find_and_load_unlocked
  File "<frozen importlib._bootstrap>", line 935, in _load_unlocked
  File "<frozen importlib._bootstrap_external>", line 999, in exec_module
  File "<frozen importlib._bootstrap>", line 488, in _call_with_frames_removed
  File "D:\anaconda3\envs\sparktts\Lib\site-packages\transformers\models\wav2vec2\modeling_wav2vec2.py", line 40, in <module>
    from ...modeling_utils import PreTrainedModel
  File "D:\anaconda3\envs\sparktts\Lib\site-packages\transformers\modeling_utils.py", line 48, in <module>
    from .loss.loss_utils import LOSS_MAPPING
  File "D:\anaconda3\envs\sparktts\Lib\site-packages\transformers\loss\loss_utils.py", line 19, in <module>
    from .loss_deformable_detr import DeformableDetrForObjectDetectionLoss, DeformableDetrForSegmentationLoss
  File "D:\anaconda3\envs\sparktts\Lib\site-packages\transformers\loss\loss_deformable_detr.py", line 4, in <module>
    from ..image_transforms import center_to_corners_format
  File "D:\anaconda3\envs\sparktts\Lib\site-packages\transformers\image_transforms.py", line 22, in <module>
    from .image_utils import (
  File "D:\anaconda3\envs\sparktts\Lib\site-packages\transformers\image_utils.py", line 58, in <module>
    from torchvision.transforms import InterpolationMode
ImportError: cannot import name 'InterpolationMode' from 'torchvision.transforms' (D:\anaconda3\envs\sparktts\Lib\site-packages\torchvision\transforms\__init__.py)

The above exception was the direct cause of the following exception:

Traceback (most recent call last):
  File "I:\AI-src\Spark-TTS\webui.py", line 23, in <module>
    from cli.SparkTTS import SparkTTS
  File "I:\AI-src\Spark-TTS\cli\SparkTTS.py", line 23, in <module>
    from sparktts.models.audio_tokenizer import BiCodecTokenizer
  File "I:\AI-src\Spark-TTS\sparktts\models\audio_tokenizer.py", line 22, in <module>
    from transformers import Wav2Vec2FeatureExtractor, Wav2Vec2Model
  File "<frozen importlib._bootstrap>", line 1412, in _handle_fromlist
  File "D:\anaconda3\envs\sparktts\Lib\site-packages\transformers\utils\import_utils.py", line 1767, in __getattr__
    value = getattr(module, name)
            ^^^^^^^^^^^^^^^^^^^^^
  File "D:\anaconda3\envs\sparktts\Lib\site-packages\transformers\utils\import_utils.py", line 1766, in __getattr__
    module = self._get_module(self._class_to_module[name])
             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\anaconda3\envs\sparktts\Lib\site-packages\transformers\utils\import_utils.py", line 1780, in _get_module
    raise RuntimeError(
RuntimeError: Failed to import transformers.models.wav2vec2.modeling_wav2vec2 because of the following error (look up to see its traceback):
cannot import name 'InterpolationMode' from 'torchvision.transforms' (D:\anaconda3\envs\sparktts\Lib\site-packages\torchvision\transforms\__init__.py)

检查 torch 和 torchvision 的版本是否匹配，可以通过以下步骤进行：

1. 查看当前安装的 torch 和 torchvision 版本

在 Python 环境中运行以下代码，获取当前安装的 torch 和 torchvision 版本：

import torch
import torchvision

print("PyTorch version:", torch.__version__)
print("TorchVision version:", torchvision.__version__)

执行后：

(sparktts) I:\AI-src\Spark-TTS>python ver.py
PyTorch version: 2.5.1+cpu
TorchVision version: 0.2.0

2. 检查版本是否匹配

根据 PyTorch 的版本，torchvision 通常会有一个对应的兼容版本。以下是一些常见版本的对应关系：

如果你的 torch 和 torchvision 版本不匹配，可能会导致兼容性问题。

3. 调整版本以确保匹配

如果发现版本不匹配，可以通过以下方法调整：

方法 1：升级或降级 torchvision

根据你的 torch 版本，安装对应的 torchvision 版本。例如：

pip install torchvision==0.20.1  # 对应 torch 2.5.x

(sparktts) I:\AI-src\Spark-TTS>pip install torchvision==0.20.1
Collecting torchvision==0.20.1
  Downloading torchvision-0.20.1-cp312-cp312-win_amd64.whl.metadata (6.2 kB)
Requirement already satisfied: numpy in d:\anaconda3\envs\sparktts\lib\site-packages (from torchvision==0.20.1) (2.2.3)
Requirement already satisfied: torch==2.5.1 in d:\anaconda3\envs\sparktts\lib\site-packages (from torchvision==0.20.1) (2.5.1)
Requirement already satisfied: pillow!=8.3.*,>=5.3.0 in d:\anaconda3\envs\sparktts\lib\site-packages (from torchvision==0.20.1) (11.1.0)
Requirement already satisfied: filelock in d:\anaconda3\envs\sparktts\lib\site-packages (from torch==2.5.1->torchvision==0.20.1) (3.17.0)
Requirement already satisfied: typing-extensions>=4.8.0 in d:\anaconda3\envs\sparktts\lib\site-packages (from torch==2.5.1->torchvision==0.20.1) (4.12.2)
Requirement already satisfied: networkx in d:\anaconda3\envs\sparktts\lib\site-packages (from torch==2.5.1->torchvision==0.20.1) (3.4.2)
Requirement already satisfied: jinja2 in d:\anaconda3\envs\sparktts\lib\site-packages (from torch==2.5.1->torchvision==0.20.1) (3.1.6)
Requirement already satisfied: fsspec in d:\anaconda3\envs\sparktts\lib\site-packages (from torch==2.5.1->torchvision==0.20.1) (2025.3.0)
Requirement already satisfied: setuptools in d:\anaconda3\envs\sparktts\lib\site-packages (from torch==2.5.1->torchvision==0.20.1) (75.8.0)
Requirement already satisfied: sympy==1.13.1 in d:\anaconda3\envs\sparktts\lib\site-packages (from torch==2.5.1->torchvision==0.20.1) (1.13.1)
Requirement already satisfied: mpmath<1.4,>=1.1.0 in d:\anaconda3\envs\sparktts\lib\site-packages (from sympy==1.13.1->torch==2.5.1->torchvision==0.20.1) (1.3.0)
Requirement already satisfied: MarkupSafe>=2.0 in d:\anaconda3\envs\sparktts\lib\site-packages (from jinja2->torch==2.5.1->torchvision==0.20.1) (2.1.5)
Downloading torchvision-0.20.1-cp312-cp312-win_amd64.whl (1.6 MB)
   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 1.6/1.6 MB 53.9 kB/s eta 0:00:00
Installing collected packages: torchvision
  Attempting uninstall: torchvision
    Found existing installation: torchvision 0.2.0
    Uninstalling torchvision-0.2.0:
      Successfully uninstalled torchvision-0.2.0
Successfully installed torchvision-0.20.1

方法 2：升级或降级 torch

如果你希望使用某个特定版本的 torchvision，可以调整 torch 的版本。例如：

pip install torch==2.4.1 torchvision==0.19.1

方法 3：使用 Conda 管理版本

如果你使用的是 Conda，可以通过以下命令安装匹配的版本：

conda install pytorch==2.4.1 torchvision==0.19.1 -c pytorch

4. 验证安装

调整版本后，重新运行以下代码，确保版本匹配且没有兼容性问题，再次验证：

(sparktts) I:\AI-src\Spark-TTS>python ver.py
PyTorch version: 2.5.1+cpu
TorchVision version: 0.20.1+cpu

如果版本匹配，但仍然遇到问题，可以尝试以下操作：

• 清理旧版本的缓存：

  pip cache purge
  

• 在一个全新的虚拟环境中重新安装依赖：

  conda create -n new_env python=3.9
  conda activate new_env
  pip install torch torchvision
  

通过上述步骤，你可以确保 torch 和 torchvision 的版本匹配，从而避免兼容性问题。

九、CUDA 运行版本对不上

PS C:\Users\jm> nvidia-smi
Mon Mar 10 22:12:53 2025
+---------------------------------------------------------------------------------------+
| NVIDIA-SMI 538.78                 Driver Version: 538.78       CUDA Version: 12.2     |
|-----------------------------------------+----------------------+----------------------+
| GPU  Name                     TCC/WDDM  | Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp   Perf          Pwr:Usage/Cap |         Memory-Usage | GPU-Util  Compute M. |
|                                         |                      |               MIG M. |
|=========================================+======================+======================|
|   0  NVIDIA GeForce MX150         WDDM  | 00000000:01:00.0 Off |                  N/A |
| N/A   50C    P0              N/A / ERR! |      0MiB /  2048MiB |      1%      Default |
|                                         |                      |                  N/A |
+-----------------------------------------+----------------------+----------------------+

+---------------------------------------------------------------------------------------+
| Processes:                                                                            |
|  GPU   GI   CI        PID   Type   Process name                            GPU Memory |
|        ID   ID                                                             Usage      |
|=======================================================================================|
|  No running processes found                                                           |
+---------------------------------------------------------------------------------------+
PS C:\Users\jm> nvcc --version
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2018 NVIDIA Corporation
Built on Sat_Aug_25_21:08:04_Central_Daylight_Time_2018
Cuda compilation tools, release 10.0, V10.0.130
PS C:\Users\jm>

显卡最高支持 12.2，但实际 CUDA 运行时版本是 10.0，需要安装个 12.2 的版本。

CUDA Toolkit Archive | NVIDIA Developer

找到对应的版本下载到本地安装。安装过程中，系统重启了，没有安装成功，重新又安装了 1 次，大约花了一小时。

C:\Users\jm>nvcc --version
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2023 NVIDIA Corporation
Built on Tue_Jun_13_19:42:34_Pacific_Daylight_Time_2023
Cuda compilation tools, release 12.2, V12.2.91
Build cuda_12.2.r12.2/compiler.32965470_0

注：根据电脑实际情况，可能有的需要安装新版本 12.1，有的只能安装旧版本 11.8。

9ong@TsingChan 文章内容由 AI 辅助生成。