jerrybwang
commited on
Commit
·
cdf446f
1
Parent(s):
92b99c9
--other 更新代码
Browse files- README.md +85 -29
- app.py +82 -340
- check_deployment.py +134 -0
- packages.txt +2 -0
- requirements.txt +3 -3
README.md
CHANGED
|
@@ -75,56 +75,112 @@ model = AutoModel.from_pretrained("FunAudioLLM/CosyVoice-300M")
|
|
| 75 |
|
| 76 |
## 部署到Hugging Face Space
|
| 77 |
|
| 78 |
-
###
|
| 79 |
|
| 80 |
1. **创建Space**
|
| 81 |
- 访问 [Hugging Face Spaces](https://huggingface.co/spaces)
|
| 82 |
- 点击 "New Space"
|
| 83 |
- 选择 Gradio SDK
|
|
|
|
| 84 |
|
| 85 |
-
2.
|
| 86 |
-
-
|
| 87 |
-
-
|
| 88 |
-
-
|
| 89 |
-
-
|
| 90 |
|
| 91 |
3. **等待构建**
|
| 92 |
-
- Space会自动安装依赖(约
|
| 93 |
-
-
|
| 94 |
- 启动应用
|
| 95 |
|
| 96 |
4. **验证部署**
|
| 97 |
- 检查模型状态指示器(应显示绿色✅)
|
| 98 |
-
-
|
|
|
|
| 99 |
|
| 100 |
-
###
|
| 101 |
|
| 102 |
-
|
| 103 |
-
|
| 104 |
-
|
|
|
|
|
|
|
| 105 |
|
| 106 |
-
###
|
| 107 |
|
| 108 |
-
|
| 109 |
-
|
| 110 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 111 |
```
|
| 112 |
|
| 113 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 114 |
|
| 115 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 116 |
|
| 117 |
-
###
|
| 118 |
-
1. 将本仓库推送到Hugging Face
|
| 119 |
-
2. 在Hugging Face网站创建新的Space
|
| 120 |
-
3. 选择"Gradio"作为SDK
|
| 121 |
-
4. 系统会自动检测配置并部署
|
| 122 |
|
| 123 |
-
|
| 124 |
-
|
| 125 |
-
-
|
| 126 |
-
-
|
| 127 |
-
-
|
|
|
|
|
|
|
| 128 |
|
| 129 |
## 技术架构
|
| 130 |
|
|
|
|
| 75 |
|
| 76 |
## 部署到Hugging Face Space
|
| 77 |
|
| 78 |
+
### 快速部署步骤
|
| 79 |
|
| 80 |
1. **创建Space**
|
| 81 |
- 访问 [Hugging Face Spaces](https://huggingface.co/spaces)
|
| 82 |
- 点击 "New Space"
|
| 83 |
- 选择 Gradio SDK
|
| 84 |
+
- 选择硬件(推荐 CPU Basic 或 GPU T4)
|
| 85 |
|
| 86 |
+
2. **上传必需文件**
|
| 87 |
+
- `app.py` - 主应用文件
|
| 88 |
+
- `requirements.txt` - Python依赖
|
| 89 |
+
- `packages.txt` - 系统依赖(ffmpeg等)
|
| 90 |
+
- `README.md` - 项目说明(包含Space配置)
|
| 91 |
|
| 92 |
3. **等待构建**
|
| 93 |
+
- Space会自动安装依赖(约5-10分钟)
|
| 94 |
+
- 自动从Hugging Face Hub下载CosyVoice-300M模型
|
| 95 |
- 启动应用
|
| 96 |
|
| 97 |
4. **验证部署**
|
| 98 |
- 检查模型状态指示器(应显示绿色✅)
|
| 99 |
+
- 测试文本转语音功能
|
| 100 |
+
- 测试音频处理功能
|
| 101 |
|
| 102 |
+
### 硬件配置建议
|
| 103 |
|
| 104 |
+
| 硬件类型 | 适用场景 | 说明 |
|
| 105 |
+
|---------|---------|------|
|
| 106 |
+
| **CPU Basic(免费)** | 测试和演示 | 推理速度较慢,适合轻量使用 |
|
| 107 |
+
| **GPU T4(推荐)** | 生产环境 | 更快的推理速度,更好的用户体验 |
|
| 108 |
+
| **GPU A10G** | 高并发 | 适合大量用户同时访问 |
|
| 109 |
|
| 110 |
+
### 重要配置说明
|
| 111 |
|
| 112 |
+
#### README.md 头部配置
|
| 113 |
+
确保 README.md 文件顶部包含以下配置:
|
| 114 |
+
```yaml
|
| 115 |
+
---
|
| 116 |
+
title: CosyVoice
|
| 117 |
+
emoji: 🌍
|
| 118 |
+
colorFrom: blue
|
| 119 |
+
colorTo: pink
|
| 120 |
+
sdk: gradio
|
| 121 |
+
sdk_version: 6.4.0
|
| 122 |
+
app_file: app.py
|
| 123 |
+
pinned: false
|
| 124 |
+
license: apache-2.0
|
| 125 |
+
---
|
| 126 |
```
|
| 127 |
|
| 128 |
+
#### 必需的文件
|
| 129 |
+
- ✅ `app.py` - 主应用
|
| 130 |
+
- ✅ `requirements.txt` - Python依赖
|
| 131 |
+
- ✅ `packages.txt` - 系统依赖(ffmpeg, libsndfile1)
|
| 132 |
+
- ✅ `README.md` - 包含Space配置
|
| 133 |
+
|
| 134 |
+
### 常见问题解决
|
| 135 |
+
|
| 136 |
+
#### 问题1: 模型加载失败
|
| 137 |
+
**症状**: 界面显示"演示模式"警告
|
| 138 |
+
**解决方案**:
|
| 139 |
+
- 检查网络连接是否正常
|
| 140 |
+
- 确认 `requirements.txt` 中包含 `transformers>=4.35.0`
|
| 141 |
+
- 查看 Space 日志确认模型下载进度
|
| 142 |
+
|
| 143 |
+
#### 问题2: 音频处理错误
|
| 144 |
+
**症状**: 上传音频后报错
|
| 145 |
+
**解决方案**:
|
| 146 |
+
- 确保 `packages.txt` 文件存在并包含 `ffmpeg`
|
| 147 |
+
- 重新构建 Space
|
| 148 |
|
| 149 |
+
#### 问题3: 依赖安装失败
|
| 150 |
+
**症状**: 构建过程中出现错误
|
| 151 |
+
**解决方案**:
|
| 152 |
+
- 检查 `requirements.txt` 格式是否正确
|
| 153 |
+
- 移除不必要的依赖(如 modelscope, edge-tts, gTTS)
|
| 154 |
+
- 使用固定版本号避免兼容性问题
|
| 155 |
+
|
| 156 |
+
### Git 部署方式
|
| 157 |
+
|
| 158 |
+
```bash
|
| 159 |
+
# 克隆或创建仓库
|
| 160 |
+
git clone https://huggingface.co/spaces/YOUR_USERNAME/YOUR_SPACE_NAME
|
| 161 |
+
cd YOUR_SPACE_NAME
|
| 162 |
+
|
| 163 |
+
# 添加文件
|
| 164 |
+
cp /path/to/app.py .
|
| 165 |
+
cp /path/to/requirements.txt .
|
| 166 |
+
cp /path/to/packages.txt .
|
| 167 |
+
cp /path/to/README.md .
|
| 168 |
+
|
| 169 |
+
# 提交并推送
|
| 170 |
+
git add .
|
| 171 |
+
git commit -m "Initial deployment"
|
| 172 |
+
git push
|
| 173 |
+
```
|
| 174 |
|
| 175 |
+
### 验证清单
|
|
|
|
|
|
|
|
|
|
|
|
|
| 176 |
|
| 177 |
+
部署前请确认:
|
| 178 |
+
- [ ] README.md 包含正确的 YAML 配置
|
| 179 |
+
- [ ] app.py 文件完整无误
|
| 180 |
+
- [ ] requirements.txt 包含所有必需依赖
|
| 181 |
+
- [ ] packages.txt 包含系统依赖
|
| 182 |
+
- [ ] 选择了合适的硬件配置
|
| 183 |
+
- [ ] Space 设置为 Public(如需公开访问)
|
| 184 |
|
| 185 |
## 技术架构
|
| 186 |
|
app.py
CHANGED
|
@@ -33,228 +33,71 @@ def load_cosyvoice_model():
|
|
| 33 |
print("="*60)
|
| 34 |
|
| 35 |
try:
|
| 36 |
-
#
|
| 37 |
-
|
| 38 |
-
|
| 39 |
-
|
| 40 |
-
import torch
|
| 41 |
-
|
| 42 |
-
model_name = "FunAudioLLM/CosyVoice-300M"
|
| 43 |
-
print(f" 从 {model_name} 加载...")
|
| 44 |
-
|
| 45 |
-
# CosyVoice需要trust_remote_code=True来加载自定义模型代码
|
| 46 |
-
model = AutoModel.from_pretrained(
|
| 47 |
-
model_name,
|
| 48 |
-
trust_remote_code=True,
|
| 49 |
-
torch_dtype=torch.float32
|
| 50 |
-
)
|
| 51 |
-
|
| 52 |
-
# 设置为评估模式
|
| 53 |
-
model.eval()
|
| 54 |
-
|
| 55 |
-
# 详细检查模型类型和方法
|
| 56 |
-
print(f" 模型类型: {type(model)}")
|
| 57 |
-
print(f" 模型类名: {model.__class__.__name__}")
|
| 58 |
-
|
| 59 |
-
# 检查所有可用的方法
|
| 60 |
-
all_methods = [m for m in dir(model) if not m.startswith('_') and callable(getattr(model, m, None))]
|
| 61 |
-
print(f" 可用方法数量: {len(all_methods)}")
|
| 62 |
-
print(f" 前20个方法: {all_methods[:20]}")
|
| 63 |
-
|
| 64 |
-
# 检查关键推理方法
|
| 65 |
-
has_inference_sft = hasattr(model, 'inference_sft')
|
| 66 |
-
has_inference = hasattr(model, 'inference')
|
| 67 |
-
has_generate = hasattr(model, 'generate')
|
| 68 |
-
has_forward = hasattr(model, 'forward')
|
| 69 |
-
|
| 70 |
-
print(f" 推理方法检查:")
|
| 71 |
-
print(f" - inference_sft: {has_inference_sft}")
|
| 72 |
-
print(f" - inference: {has_inference}")
|
| 73 |
-
print(f" - generate: {has_generate}")
|
| 74 |
-
print(f" - forward: {has_forward}")
|
| 75 |
-
|
| 76 |
-
# 如果模型有这些方法,说明加载成功
|
| 77 |
-
has_inference = has_inference_sft or has_inference or has_generate
|
| 78 |
-
|
| 79 |
-
if has_inference:
|
| 80 |
-
cosyvoice_model = {
|
| 81 |
-
'model': model,
|
| 82 |
-
'type': 'transformers',
|
| 83 |
-
'has_inference': True
|
| 84 |
-
}
|
| 85 |
-
model_loaded = True
|
| 86 |
-
print(" ✓ 成功通过transformers加载CosyVoice模型(有推理方法)")
|
| 87 |
-
print("="*60 + "\n")
|
| 88 |
-
return cosyvoice_model
|
| 89 |
-
else:
|
| 90 |
-
print(" ⚠ 模型加载成功但缺少推理方法,尝试下一个方法...")
|
| 91 |
-
# 不返回,继续尝试其他方法
|
| 92 |
-
|
| 93 |
-
except Exception as e:
|
| 94 |
-
print(f" ✗ transformers加载失败: {e}")
|
| 95 |
-
import traceback
|
| 96 |
-
print(f" 详细错误: {traceback.format_exc()[:300]}")
|
| 97 |
|
| 98 |
-
|
| 99 |
-
|
| 100 |
-
print("\n[方法2] 尝试从Hugging Face Hub下载并使用官方CosyVoice...")
|
| 101 |
-
from huggingface_hub import snapshot_download
|
| 102 |
-
import torch
|
| 103 |
-
|
| 104 |
-
model_id = "FunAudioLLM/CosyVoice-300M"
|
| 105 |
-
print(f" 下载模型: {model_id}")
|
| 106 |
-
|
| 107 |
-
model_dir = snapshot_download(
|
| 108 |
-
repo_id=model_id,
|
| 109 |
-
allow_patterns=["*.pt", "*.pth", "*.bin", "*.safetensors", "*.json", "*.txt", "*.yaml", "*.py"],
|
| 110 |
-
)
|
| 111 |
-
print(f" 模型下载到: {model_dir}")
|
| 112 |
-
|
| 113 |
-
# 尝试导入CosyVoice官方包
|
| 114 |
-
try:
|
| 115 |
-
# 首先尝试直接导入(如果已安装)
|
| 116 |
-
from cosyvoice.cli.cosyvoice import CosyVoice
|
| 117 |
-
print(" 使用已安装的CosyVoice包")
|
| 118 |
-
cosyvoice_model = CosyVoice(model_dir)
|
| 119 |
-
model_loaded = True
|
| 120 |
-
print(" ✓ 成功使用官方CosyVoice包加载模型")
|
| 121 |
-
print("="*60 + "\n")
|
| 122 |
-
return cosyvoice_model
|
| 123 |
-
except ImportError:
|
| 124 |
-
print(" CosyVoice包未安装,尝试从下载的代码加载...")
|
| 125 |
-
|
| 126 |
-
# 尝试从下载的模型目录加载代码
|
| 127 |
-
if model_dir not in sys.path:
|
| 128 |
-
sys.path.insert(0, model_dir)
|
| 129 |
-
|
| 130 |
-
# 查找并加载modeling文件
|
| 131 |
-
import glob
|
| 132 |
-
py_files = glob.glob(os.path.join(model_dir, "**/*.py"), recursive=True)
|
| 133 |
-
print(f" 找到Python文件: {len(py_files)}个")
|
| 134 |
-
|
| 135 |
-
# 尝试找到CosyVoice类定义
|
| 136 |
-
for py_file in py_files:
|
| 137 |
-
if 'cosyvoice' in py_file.lower() or 'model' in py_file.lower():
|
| 138 |
-
print(f" 检查文件: {os.path.basename(py_file)}")
|
| 139 |
-
|
| 140 |
-
# 如果找不到官方包,尝试直接加载模型权重
|
| 141 |
-
try:
|
| 142 |
-
# 查找模型文件
|
| 143 |
-
model_files = glob.glob(os.path.join(model_dir, "**/*.pt"), recursive=True)
|
| 144 |
-
model_files += glob.glob(os.path.join(model_dir, "**/*.pth"), recursive=True)
|
| 145 |
-
model_files += glob.glob(os.path.join(model_dir, "**/*.bin"), recursive=True)
|
| 146 |
-
|
| 147 |
-
if model_files:
|
| 148 |
-
print(f" 找到模型文件: {len(model_files)}个")
|
| 149 |
-
for mf in model_files[:5]:
|
| 150 |
-
print(f" - {os.path.basename(mf)}")
|
| 151 |
-
|
| 152 |
-
# 尝试使用transformers的AutoModel加载
|
| 153 |
-
print(" 尝试使用AutoModel从本地目录加载...")
|
| 154 |
-
from transformers import AutoModel
|
| 155 |
-
model = AutoModel.from_pretrained(
|
| 156 |
-
model_dir,
|
| 157 |
-
trust_remote_code=True,
|
| 158 |
-
local_files_only=True,
|
| 159 |
-
torch_dtype=torch.float32
|
| 160 |
-
)
|
| 161 |
-
model.eval()
|
| 162 |
-
|
| 163 |
-
# 检查推理方法
|
| 164 |
-
print(f" 模型类型: {type(model).__name__}")
|
| 165 |
-
has_inference = hasattr(model, 'inference_sft') or hasattr(model, 'inference') or hasattr(model, 'generate')
|
| 166 |
-
print(f" 有推理方法: {has_inference}")
|
| 167 |
-
|
| 168 |
-
if has_inference:
|
| 169 |
-
cosyvoice_model = {
|
| 170 |
-
'model': model,
|
| 171 |
-
'model_dir': model_dir,
|
| 172 |
-
'type': 'transformers'
|
| 173 |
-
}
|
| 174 |
-
model_loaded = True
|
| 175 |
-
print(" ✓ 成功从本地目录加载模型")
|
| 176 |
-
print("="*60 + "\n")
|
| 177 |
-
return cosyvoice_model
|
| 178 |
-
else:
|
| 179 |
-
print(" ⚠ 模型加载成功但缺少推理方法")
|
| 180 |
-
else:
|
| 181 |
-
print(" ✗ 未找到模型文件")
|
| 182 |
-
|
| 183 |
-
except Exception as load_err:
|
| 184 |
-
print(f" ✗ 加载模型文件失败: {load_err}")
|
| 185 |
-
import traceback
|
| 186 |
-
print(f" 详细错误: {traceback.format_exc()[:300]}")
|
| 187 |
-
|
| 188 |
-
except Exception as e:
|
| 189 |
-
print(f" ✗ Hugging Face Hub下载失败: {e}")
|
| 190 |
-
import traceback
|
| 191 |
-
print(f" 详细错误: {traceback.format_exc()[:300]}")
|
| 192 |
|
| 193 |
-
#
|
| 194 |
-
|
| 195 |
-
|
| 196 |
-
|
| 197 |
-
|
| 198 |
-
|
| 199 |
-
|
| 200 |
-
'pretrained_models/CosyVoice-300M',
|
| 201 |
-
'CosyVoice-300M',
|
| 202 |
-
'./models/CosyVoice-300M',
|
| 203 |
-
]
|
| 204 |
-
|
| 205 |
-
for model_dir in possible_paths:
|
| 206 |
-
if model_dir and os.path.exists(model_dir):
|
| 207 |
-
print(f" 尝试从路径加载: {model_dir}")
|
| 208 |
-
try:
|
| 209 |
-
cosyvoice_model = CosyVoice(model_dir)
|
| 210 |
-
model_loaded = True
|
| 211 |
-
print(f" ✓ 成功从 {model_dir} 加载CosyVoice模型")
|
| 212 |
-
print("="*60 + "\n")
|
| 213 |
-
return cosyvoice_model
|
| 214 |
-
except Exception as e:
|
| 215 |
-
print(f" ✗ 加载失败: {e}")
|
| 216 |
-
continue
|
| 217 |
-
|
| 218 |
-
except ImportError as e:
|
| 219 |
-
print(f" ✗ CosyVoice包未安装: {e}")
|
| 220 |
-
except Exception as e:
|
| 221 |
-
print(f" ✗ 方法3失败: {e}")
|
| 222 |
|
| 223 |
-
#
|
| 224 |
-
|
| 225 |
-
|
| 226 |
-
|
| 227 |
-
|
| 228 |
-
|
| 229 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 230 |
|
| 231 |
-
|
| 232 |
-
|
| 233 |
-
|
| 234 |
-
|
| 235 |
-
cosyvoice_model = CosyVoice(model_dir)
|
| 236 |
-
model_loaded = True
|
| 237 |
-
print(" ✓ 成功通过ModelScope加载CosyVoice模型")
|
| 238 |
-
print("="*60 + "\n")
|
| 239 |
-
return cosyvoice_model
|
| 240 |
-
except ImportError:
|
| 241 |
-
cosyvoice_model = {'model_dir': model_dir, 'type': 'downloaded'}
|
| 242 |
-
model_loaded = True
|
| 243 |
-
print(" ✓ 模型文件已下载(演示模式)")
|
| 244 |
-
print("="*60 + "\n")
|
| 245 |
-
return cosyvoice_model
|
| 246 |
-
except Exception as e:
|
| 247 |
-
print(f" ✗ ModelScope加载失败: {e}")
|
| 248 |
|
| 249 |
-
#
|
| 250 |
-
print("\n
|
| 251 |
-
print("
|
| 252 |
-
print("
|
| 253 |
-
print("
|
| 254 |
-
print("
|
| 255 |
-
print(" 3. transformers包版本 >= 4.35.0")
|
| 256 |
-
print("\n安装方法:")
|
| 257 |
-
print(" pip install transformers>=4.35.0")
|
| 258 |
print("="*60 + "\n")
|
| 259 |
|
| 260 |
cosyvoice_model = None
|
|
@@ -497,129 +340,28 @@ def text_to_speech(text, speaker="中文女"):
|
|
| 497 |
return None, f"语音合成失败: {str(e)}"
|
| 498 |
|
| 499 |
def generate_demo_audio(text, speaker, error=None):
|
| 500 |
-
"""
|
| 501 |
-
|
| 502 |
-
|
| 503 |
-
|
| 504 |
-
|
| 505 |
-
|
| 506 |
-
|
| 507 |
-
|
| 508 |
-
|
| 509 |
-
|
| 510 |
-
|
| 511 |
-
|
| 512 |
-
|
| 513 |
-
|
| 514 |
-
|
| 515 |
-
|
| 516 |
-
|
| 517 |
-
|
| 518 |
-
|
| 519 |
-
|
| 520 |
-
|
| 521 |
-
|
| 522 |
-
# 异步生成语音
|
| 523 |
-
async def generate():
|
| 524 |
-
communicate = edge_tts.Communicate(text, voice)
|
| 525 |
-
await communicate.save(tmp_path)
|
| 526 |
-
|
| 527 |
-
# 运行异步任务
|
| 528 |
-
asyncio.run(generate())
|
| 529 |
-
|
| 530 |
-
# 读取生成的音频
|
| 531 |
-
import soundfile as sf
|
| 532 |
-
from pydub import AudioSegment
|
| 533 |
-
|
| 534 |
-
# 转换MP3到WAV
|
| 535 |
-
audio = AudioSegment.from_mp3(tmp_path)
|
| 536 |
-
wav_path = tmp_path.replace('.mp3', '.wav')
|
| 537 |
-
audio.export(wav_path, format='wav')
|
| 538 |
-
|
| 539 |
-
# 读取WAV文件
|
| 540 |
-
audio_data, sample_rate = sf.read(wav_path)
|
| 541 |
-
|
| 542 |
-
# 清理临时文件
|
| 543 |
-
os.unlink(tmp_path)
|
| 544 |
-
os.unlink(wav_path)
|
| 545 |
-
|
| 546 |
-
audio_tuple = (sample_rate, audio_data.astype(np.float32))
|
| 547 |
-
|
| 548 |
-
status_msg = f"✓ 语音合成成功(使用Edge TTS)\n文本: {text}\n说话人: {speaker}\n"
|
| 549 |
-
if error:
|
| 550 |
-
status_msg += f"注意: CosyVoice模型不可用,已使用备用引擎\n"
|
| 551 |
-
|
| 552 |
-
print(f" ✓ Edge TTS生成成功")
|
| 553 |
-
return audio_tuple, status_msg
|
| 554 |
-
|
| 555 |
-
except Exception as e:
|
| 556 |
-
print(f" ✗ Edge TTS失败: {e}")
|
| 557 |
-
|
| 558 |
-
# 如果edge-tts失败,尝试gTTS
|
| 559 |
-
try:
|
| 560 |
-
from gtts import gTTS
|
| 561 |
-
import tempfile
|
| 562 |
-
import soundfile as sf
|
| 563 |
-
from pydub import AudioSegment
|
| 564 |
-
|
| 565 |
-
print(f"使用gTTS生成语音: {text[:50]}...")
|
| 566 |
-
|
| 567 |
-
# 根据说话人选择语言
|
| 568 |
-
lang = 'zh-CN' if speaker.startswith('中文') else 'en'
|
| 569 |
-
|
| 570 |
-
# 创建临时文件
|
| 571 |
-
with tempfile.NamedTemporaryFile(delete=False, suffix='.mp3') as tmp_file:
|
| 572 |
-
tmp_path = tmp_file.name
|
| 573 |
-
|
| 574 |
-
# 生成语音
|
| 575 |
-
tts = gTTS(text=text, lang=lang, slow=False)
|
| 576 |
-
tts.save(tmp_path)
|
| 577 |
-
|
| 578 |
-
# 转换MP3到WAV
|
| 579 |
-
audio = AudioSegment.from_mp3(tmp_path)
|
| 580 |
-
wav_path = tmp_path.replace('.mp3', '.wav')
|
| 581 |
-
audio.export(wav_path, format='wav')
|
| 582 |
-
|
| 583 |
-
# 读取WAV文件
|
| 584 |
-
audio_data, sample_rate = sf.read(wav_path)
|
| 585 |
-
|
| 586 |
-
# 清理临时文件
|
| 587 |
-
os.unlink(tmp_path)
|
| 588 |
-
os.unlink(wav_path)
|
| 589 |
-
|
| 590 |
-
audio_tuple = (sample_rate, audio_data.astype(np.float32))
|
| 591 |
-
|
| 592 |
-
status_msg = f"✓ 语音合成成功(使用Google TTS)\n文本: {text}\n说话人: {speaker}\n"
|
| 593 |
-
if error:
|
| 594 |
-
status_msg += f"注意: CosyVoice模型不可用,已使用备用引擎\n"
|
| 595 |
-
|
| 596 |
-
print(f" ✓ gTTS生成成功")
|
| 597 |
-
return audio_tuple, status_msg
|
| 598 |
-
|
| 599 |
-
except Exception as e2:
|
| 600 |
-
print(f" ✗ gTTS也失败: {e2}")
|
| 601 |
-
|
| 602 |
-
# 最后的降级方案:生成演示音频
|
| 603 |
-
sample_rate = 22050
|
| 604 |
-
duration = min(len(text) * 0.2, 5.0)
|
| 605 |
-
t = np.linspace(0, duration, int(sample_rate * duration), False)
|
| 606 |
-
|
| 607 |
-
frequency = 440
|
| 608 |
-
audio_data = 0.3 * np.sin(2 * np.pi * frequency * t)
|
| 609 |
-
audio_data += 0.2 * np.sin(2 * np.pi * frequency * 1.5 * t)
|
| 610 |
-
|
| 611 |
-
fade_samples = int(sample_rate * 0.1)
|
| 612 |
-
audio_data[:fade_samples] *= np.linspace(0, 1, fade_samples)
|
| 613 |
-
audio_data[-fade_samples:] *= np.linspace(1, 0, fade_samples)
|
| 614 |
-
|
| 615 |
-
audio_tuple = (sample_rate, audio_data.astype(np.float32))
|
| 616 |
-
|
| 617 |
-
status_msg = f"⚠ 演示模式\n文本: {text}\n说话人: {speaker}\n"
|
| 618 |
-
if error:
|
| 619 |
-
status_msg += f"注意: 所有TTS引擎都不可用\n"
|
| 620 |
-
status_msg += "提示: 这是演示音频,不是真实的语音合成结果"
|
| 621 |
-
|
| 622 |
-
return audio_tuple, status_msg
|
| 623 |
|
| 624 |
# 在启动时加载模型
|
| 625 |
load_cosyvoice_model()
|
|
|
|
| 33 |
print("="*60)
|
| 34 |
|
| 35 |
try:
|
| 36 |
+
# 使用transformers加载(推荐用于Hugging Face Space)
|
| 37 |
+
print("\n尝试使用transformers加载...")
|
| 38 |
+
from transformers import AutoModel
|
| 39 |
+
import torch
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 40 |
|
| 41 |
+
model_name = "FunAudioLLM/CosyVoice-300M"
|
| 42 |
+
print(f"从 {model_name} 加载...")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 43 |
|
| 44 |
+
# CosyVoice需要trust_remote_code=True来加载自定义模型代码
|
| 45 |
+
model = AutoModel.from_pretrained(
|
| 46 |
+
model_name,
|
| 47 |
+
trust_remote_code=True,
|
| 48 |
+
torch_dtype=torch.float32,
|
| 49 |
+
low_cpu_mem_usage=True
|
| 50 |
+
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 51 |
|
| 52 |
+
# 设置为评估模式
|
| 53 |
+
model.eval()
|
| 54 |
+
|
| 55 |
+
# 检查模型类型和方法
|
| 56 |
+
print(f"模型类型: {type(model).__name__}")
|
| 57 |
+
|
| 58 |
+
# 检查关键推理方法
|
| 59 |
+
has_inference_sft = hasattr(model, 'inference_sft')
|
| 60 |
+
has_inference = hasattr(model, 'inference')
|
| 61 |
+
has_generate = hasattr(model, 'generate')
|
| 62 |
+
|
| 63 |
+
print(f"推理方法检查:")
|
| 64 |
+
print(f" - inference_sft: {has_inference_sft}")
|
| 65 |
+
print(f" - inference: {has_inference}")
|
| 66 |
+
print(f" - generate: {has_generate}")
|
| 67 |
+
|
| 68 |
+
# 如果模型有这些方法,说明加载成功
|
| 69 |
+
if has_inference_sft or has_inference or has_generate:
|
| 70 |
+
cosyvoice_model = {
|
| 71 |
+
'model': model,
|
| 72 |
+
'type': 'transformers',
|
| 73 |
+
'has_inference': True
|
| 74 |
+
}
|
| 75 |
+
model_loaded = True
|
| 76 |
+
print("✓ 成功加载CosyVoice模型")
|
| 77 |
+
print("="*60 + "\n")
|
| 78 |
+
return cosyvoice_model
|
| 79 |
+
else:
|
| 80 |
+
print("⚠ 模型加载成功但缺少推理方法")
|
| 81 |
+
cosyvoice_model = {
|
| 82 |
+
'model': model,
|
| 83 |
+
'type': 'transformers',
|
| 84 |
+
'has_inference': False
|
| 85 |
+
}
|
| 86 |
+
model_loaded = True
|
| 87 |
+
print("="*60 + "\n")
|
| 88 |
+
return cosyvoice_model
|
| 89 |
|
| 90 |
+
except Exception as e:
|
| 91 |
+
print(f"✗ 模型加载失败: {e}")
|
| 92 |
+
import traceback
|
| 93 |
+
print(f"详细错误:\n{traceback.format_exc()}")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 94 |
|
| 95 |
+
# 演示模式(加载失败)
|
| 96 |
+
print("\n⚠ 使用演示模式")
|
| 97 |
+
print("提示: 要使用完整功能,请确保:")
|
| 98 |
+
print(" 1. 网络连接正常,可以访问Hugging Face")
|
| 99 |
+
print(" 2. 有足够的磁盘空间(约2GB)")
|
| 100 |
+
print(" 3. transformers包版本 >= 4.35.0")
|
|
|
|
|
|
|
|
|
|
| 101 |
print("="*60 + "\n")
|
| 102 |
|
| 103 |
cosyvoice_model = None
|
|
|
|
| 340 |
return None, f"语音合成失败: {str(e)}"
|
| 341 |
|
| 342 |
def generate_demo_audio(text, speaker, error=None):
|
| 343 |
+
"""生成演示音频(当模型不可用时)"""
|
| 344 |
+
# 生成简单的演示音频
|
| 345 |
+
sample_rate = 22050
|
| 346 |
+
duration = min(len(text) * 0.2, 5.0)
|
| 347 |
+
t = np.linspace(0, duration, int(sample_rate * duration), False)
|
| 348 |
+
|
| 349 |
+
frequency = 440
|
| 350 |
+
audio_data = 0.3 * np.sin(2 * np.pi * frequency * t)
|
| 351 |
+
audio_data += 0.2 * np.sin(2 * np.pi * frequency * 1.5 * t)
|
| 352 |
+
|
| 353 |
+
fade_samples = int(sample_rate * 0.1)
|
| 354 |
+
audio_data[:fade_samples] *= np.linspace(0, 1, fade_samples)
|
| 355 |
+
audio_data[-fade_samples:] *= np.linspace(1, 0, fade_samples)
|
| 356 |
+
|
| 357 |
+
audio_tuple = (sample_rate, audio_data.astype(np.float32))
|
| 358 |
+
|
| 359 |
+
status_msg = f"⚠ 演示模式\n文本: {text}\n说话人: {speaker}\n"
|
| 360 |
+
if error:
|
| 361 |
+
status_msg += f"错误: {error}\n"
|
| 362 |
+
status_msg += "提示: 这是演示音频,不是真实的语音合成结果。请确保模型正确加载。"
|
| 363 |
+
|
| 364 |
+
return audio_tuple, status_msg
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 365 |
|
| 366 |
# 在启动时加载模型
|
| 367 |
load_cosyvoice_model()
|
check_deployment.py
ADDED
|
@@ -0,0 +1,134 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
#!/usr/bin/env python3
|
| 2 |
+
"""
|
| 3 |
+
Hugging Face Space 部署检查脚本
|
| 4 |
+
检查所有必需的文件和配置是否正确
|
| 5 |
+
"""
|
| 6 |
+
|
| 7 |
+
import os
|
| 8 |
+
import sys
|
| 9 |
+
from pathlib import Path
|
| 10 |
+
|
| 11 |
+
def check_file_exists(filepath, required=True):
|
| 12 |
+
"""检查文件是否存在"""
|
| 13 |
+
exists = os.path.exists(filepath)
|
| 14 |
+
status = "✅" if exists else ("❌" if required else "⚠️")
|
| 15 |
+
req_text = "必需" if required else "可选"
|
| 16 |
+
print(f"{status} {filepath} ({req_text}): {'存在' if exists else '缺失'}")
|
| 17 |
+
return exists
|
| 18 |
+
|
| 19 |
+
def check_file_content(filepath, required_content):
|
| 20 |
+
"""检查文件内容"""
|
| 21 |
+
try:
|
| 22 |
+
with open(filepath, 'r', encoding='utf-8') as f:
|
| 23 |
+
content = f.read()
|
| 24 |
+
for item in required_content:
|
| 25 |
+
if item in content:
|
| 26 |
+
print(f" ✅ 包含: {item}")
|
| 27 |
+
else:
|
| 28 |
+
print(f" ❌ 缺失: {item}")
|
| 29 |
+
return False
|
| 30 |
+
return True
|
| 31 |
+
except Exception as e:
|
| 32 |
+
print(f" ❌ 读取文件失败: {e}")
|
| 33 |
+
return False
|
| 34 |
+
|
| 35 |
+
def main():
|
| 36 |
+
print("="*60)
|
| 37 |
+
print("Hugging Face Space 部署检查")
|
| 38 |
+
print("="*60)
|
| 39 |
+
print()
|
| 40 |
+
|
| 41 |
+
# 检查必需文件
|
| 42 |
+
print("📋 检查必需文件:")
|
| 43 |
+
print("-"*60)
|
| 44 |
+
|
| 45 |
+
files_ok = True
|
| 46 |
+
files_ok &= check_file_exists("app.py", required=True)
|
| 47 |
+
files_ok &= check_file_exists("requirements.txt", required=True)
|
| 48 |
+
files_ok &= check_file_exists("packages.txt", required=True)
|
| 49 |
+
files_ok &= check_file_exists("README.md", required=True)
|
| 50 |
+
|
| 51 |
+
print()
|
| 52 |
+
|
| 53 |
+
# 检查可选文件
|
| 54 |
+
print("📋 检查可选文件:")
|
| 55 |
+
print("-"*60)
|
| 56 |
+
check_file_exists("config.py", required=False)
|
| 57 |
+
check_file_exists(".gitignore", required=False)
|
| 58 |
+
|
| 59 |
+
print()
|
| 60 |
+
|
| 61 |
+
# 检查 README.md 配置
|
| 62 |
+
print("📋 检查 README.md 配置:")
|
| 63 |
+
print("-"*60)
|
| 64 |
+
if os.path.exists("README.md"):
|
| 65 |
+
readme_items = [
|
| 66 |
+
"title:",
|
| 67 |
+
"sdk: gradio",
|
| 68 |
+
"app_file: app.py",
|
| 69 |
+
]
|
| 70 |
+
check_file_content("README.md", readme_items)
|
| 71 |
+
|
| 72 |
+
print()
|
| 73 |
+
|
| 74 |
+
# 检查 requirements.txt
|
| 75 |
+
print("📋 检查 requirements.txt:")
|
| 76 |
+
print("-"*60)
|
| 77 |
+
if os.path.exists("requirements.txt"):
|
| 78 |
+
req_items = [
|
| 79 |
+
"gradio",
|
| 80 |
+
"torch",
|
| 81 |
+
"transformers",
|
| 82 |
+
"huggingface_hub",
|
| 83 |
+
]
|
| 84 |
+
check_file_content("requirements.txt", req_items)
|
| 85 |
+
|
| 86 |
+
print()
|
| 87 |
+
|
| 88 |
+
# 检查 packages.txt
|
| 89 |
+
print("📋 检查 packages.txt:")
|
| 90 |
+
print("-"*60)
|
| 91 |
+
if os.path.exists("packages.txt"):
|
| 92 |
+
pkg_items = [
|
| 93 |
+
"ffmpeg",
|
| 94 |
+
]
|
| 95 |
+
check_file_content("packages.txt", pkg_items)
|
| 96 |
+
|
| 97 |
+
print()
|
| 98 |
+
|
| 99 |
+
# 检查 app.py
|
| 100 |
+
print("📋 检查 app.py:")
|
| 101 |
+
print("-"*60)
|
| 102 |
+
if os.path.exists("app.py"):
|
| 103 |
+
app_items = [
|
| 104 |
+
"import gradio",
|
| 105 |
+
"AutoModel.from_pretrained",
|
| 106 |
+
"FunAudioLLM/CosyVoice-300M",
|
| 107 |
+
"demo.launch()",
|
| 108 |
+
]
|
| 109 |
+
check_file_content("app.py", app_items)
|
| 110 |
+
|
| 111 |
+
print()
|
| 112 |
+
print("="*60)
|
| 113 |
+
|
| 114 |
+
if files_ok:
|
| 115 |
+
print("✅ 所有必需文件检查通过!")
|
| 116 |
+
print()
|
| 117 |
+
print("📦 下一步:")
|
| 118 |
+
print("1. 访问 https://huggingface.co/spaces")
|
| 119 |
+
print("2. 创建新的 Space,选择 Gradio SDK")
|
| 120 |
+
print("3. 上传以下文件:")
|
| 121 |
+
print(" - app.py")
|
| 122 |
+
print(" - requirements.txt")
|
| 123 |
+
print(" - packages.txt")
|
| 124 |
+
print(" - README.md")
|
| 125 |
+
print("4. 等待构建完成(约5-10分钟)")
|
| 126 |
+
print("5. 测试应用功能")
|
| 127 |
+
else:
|
| 128 |
+
print("❌ 检查失败!请修复上述问题后重试。")
|
| 129 |
+
sys.exit(1)
|
| 130 |
+
|
| 131 |
+
print("="*60)
|
| 132 |
+
|
| 133 |
+
if __name__ == "__main__":
|
| 134 |
+
main()
|
packages.txt
ADDED
|
@@ -0,0 +1,2 @@
|
|
|
|
|
|
|
|
|
|
| 1 |
+
ffmpeg
|
| 2 |
+
libsndfile1
|
requirements.txt
CHANGED
|
@@ -7,7 +7,7 @@ librosa>=0.10.0
|
|
| 7 |
soundfile>=0.12.0
|
| 8 |
scipy>=1.10.0
|
| 9 |
huggingface_hub>=0.19.0
|
| 10 |
-
modelscope
|
| 11 |
pydub
|
| 12 |
-
|
| 13 |
-
|
|
|
|
|
|
| 7 |
soundfile>=0.12.0
|
| 8 |
scipy>=1.10.0
|
| 9 |
huggingface_hub>=0.19.0
|
|
|
|
| 10 |
pydub
|
| 11 |
+
accelerate
|
| 12 |
+
sentencepiece
|
| 13 |
+
protobuf
|