Upload 9 files

.gitattributes

@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+build/visqol/visqol_lib_py.so filter=lfs diff=lfs merge=lfs -text

Dockerfile

@@ -0,0 +1,36 @@
+# 使用官方 Python 3.8 镜像以匹配 .so 文件
+FROM python:3.8-slim
+
+# 设置工作目录
+WORKDIR /app
+
+# 安装系统依赖 (libsndfile1 用于 soundfile, ffmpeg 用于转换)
+RUN apt-get update && apt-get install -y --no-install-recommends \
+    libsndfile1 \
+    ffmpeg \
+    && apt-get clean \
+    && rm -rf /var/lib/apt/lists/*
+
+# 复制 ViSQOL 构建文件到容器中的 /app/build 目录下
+# 注意：源路径相对于 Dockerfile 所在位置
+COPY ./build /app/build
+
+# 复制项目文件到容器中
+COPY requirements.txt app.py ./
+
+# 安装 Python 依赖
+RUN pip install --no-cache-dir -r requirements.txt
+
+# 将 ViSQOL 库所在的目录添加到动态链接器查找路径
+# 这样 Python 的 ctypes 或 CFFI 才能找到 .so 文件
+ENV LD_LIBRARY_PATH=/app/build/visqol:${LD_LIBRARY_PATH}
+
+# 确保 ViSQOL 库有执行权限 (虽然通常不需要对 .so 设置执行权限，但以防万一)
+# RUN chmod +x /app/build/visqol/visqol_lib_py.so
+
+# 暴露 FastAPI 默认使用的端口 (虽然 HF Spaces 会处理端口映射)
+EXPOSE 8000
+
+# 启动 FastAPI 应用
+# 使用 uvicorn 运行 app.py 中的 app 实例
+CMD ["uvicorn", "app:app", "--host", "0.0.0.0", "--port", "8000"] 

README.md

@@ -1,11 +1,60 @@
----
-title: Visqol
-emoji: 📊
-colorFrom: gray
-colorTo: pink
-sdk: docker
-pinned: false
-license: mit
----
-
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+---
+title: ViSQOL Audio Quality API
+emoji: 🎧
+colorFrom: blue
+colorTo: green
+sdk: docker
+app_port: 8000
+# pinned: false
+# license: apache-2.0 # 如果你想指定许可证
+---
+
+# ViSQOL 音频质量评估 API
+
+这是一个基于 FastAPI 的 Hugging Face Space，用于提供 ViSQOL 音频质量评估服务。
+
+它使用了 Google 的 ViSQOL 算法 (Linux 编译版本，适用于 Python 3.8) 来计算参考音频和待评估音频之间的感知相似度得分 (MOS-LQO)。
+
+## API 端点
+
+*   **`POST /evaluate/`**
+    *   接收两个音频文件 (`reference` 和 `degraded`) 以及一个模式参数 (`mode`, 'audio' 或 'speech')。
+    *   返回包含 MOS-LQO 得分和其他信息的 JSON 响应。
+
+## 如何使用
+
+你可以通过发送 POST 请求到部署后的 Space URL 的 `/evaluate/` 路径来使用此 API。
+
+**示例 (Python):**
+
+```python
+import requests
+
+# 替换为你的 Space URL
+API_URL = "https://你的用户名-你的spacename.hf.space/evaluate/" 
+
+files = {
+    'reference': open('path/to/reference.wav', 'rb'),
+    'degraded': open('path/to/degraded.wav', 'rb')
+}
+
+params = {'mode': 'audio'} # 或 'speech'
+
+response = requests.post(API_URL, files=files, params=params)
+
+if response.status_code == 200:
+    result = response.json()
+    print(f"评估结果: {result}")
+    if result['status'] == '处理成功':
+        print(f"MOS-LQO: {result['moslqo']}")
+    else:
+        print(f"处理失败: {result['error_message']}")
+else:
+    print(f"API 请求错误: {response.status_code} - {response.text}")
+```
+
+## 注意
+
+*   输入的音频文件推荐使用 WAV 格式。
+*   参考音频和待评估音频的采样率应该匹配。
+*   此 Space 使用的 ViSQOL 库是为 Python 3.8 编译的。 

app.py

@@ -0,0 +1,263 @@
+from fastapi import FastAPI, UploadFile, File, HTTPException
+from fastapi.responses import JSONResponse
+import subprocess
+import tempfile
+import os
+import shutil
+from pydantic import BaseModel
+import sys
+import numpy as np # ViSQOL 可能需要 numpy
+import soundfile as sf # 用于读取音频
+from typing import Optional, List # 导入 List
+import librosa # Need librosa for resampling during conversion if soundfile fails
+
+app = FastAPI(title="ViSQOL 音频质量 API")
+
+# --- 配置 ViSQOL 路径 ---
+# 相对于 app.py 的路径
+VISQOL_DIR = "./build/visqol"
+VISQOL_LIB_PATH = os.path.join(VISQOL_DIR, "visqol_lib_py.so")
+PB2_DIR = os.path.join(VISQOL_DIR, "pb2") # pb2 文件所在的目录
+MODEL_DIR = os.path.join(VISQOL_DIR, "model")
+SPEECH_MODEL_PATH = os.path.join(MODEL_DIR, "libsvm_nu_svr_model.txt")
+AUDIO_MODEL_PATH = os.path.join(MODEL_DIR, "lattice_tcditugenmeetpackhref_ls2_nl60_lr12_bs2048_learn.005_ep2400_train1_7_raw.tflite")
+# --- 路径配置结束 ---
+
+# 检查文件是否存在
+required_files = [VISQOL_LIB_PATH, SPEECH_MODEL_PATH, AUDIO_MODEL_PATH]
+if not all(os.path.exists(f) for f in required_files):
+    missing = [f for f in required_files if not os.path.exists(f)]
+    raise FileNotFoundError(f"ViSQOL 必需文件未找到: {', '.join(missing)}")
+if not os.path.exists(PB2_DIR) or not os.path.isdir(PB2_DIR):
+     raise FileNotFoundError(f"ViSQOL pb2 目录未找到: {PB2_DIR}")
+
+# 动态导入 ViSQOL 库和 pb2 文件
+try:
+    # 将 pb2 目录和 visqol 目录添加到 Python 路径
+    sys.path.insert(0, os.path.abspath(PB2_DIR))
+    sys.path.insert(0, os.path.abspath(VISQOL_DIR))
+    # 加载 .so 文件需要确保 Python 能找到它，或者它在 LD_LIBRARY_PATH 中
+    # 通常放在 sys.path 中对于纯 Python 导入是足够的，但 .so 可能不同
+    # 在 Dockerfile 中我们会处理库路径
+    import visqol_lib_py
+    import similarity_result_pb2
+    import visqol_config_pb2
+    print("ViSQOL 库和 pb2 文件导入成功。")
+except ImportError as e:
+    print(f"错误：无法导入 ViSQOL 库或 pb2 文件。")
+    print(f"Python 搜索路径: {sys.path}")
+    print(f"错误详情: {e}")
+    # 在 Hugging Face 环境中，启动失败会显示日志，所以这里不直接 raise
+    # raise ImportError(f"无法导入 ViSQOL 库或 pb2 文件: {e}")
+    visqol_lib_py = None # 标记为不可用
+
+# 定义 API 响应模型
+class VisqolResponse(BaseModel):
+    reference_filename: str
+    degraded_filename: str
+    mode: str
+    moslqo: float
+    vnsim: Optional[float] = None # 添加 vnsim 字段，设为可选
+    fvnsim: Optional[List[float]] = None # 添加 fvnsim 字段，设为可选
+    status: str
+    error_message: Optional[str] = None
+
+# Function to convert and resample audio using ffmpeg
+def convert_and_resample_audio(input_path, output_path, target_sr):
+    """Converts audio to WAV format and resamples using ffmpeg."""
+    cmd = [
+        'ffmpeg',
+        '-y', # Overwrite output file if it exists
+        '-i', input_path,
+        '-ar', str(target_sr), # Set target sample rate
+        '-ac', '1', # Force mono channel (ViSQOL often expects mono)
+        output_path
+    ]
+    print(f"Running ffmpeg: {' '.join(cmd)}")
+    try:
+        result = subprocess.run(cmd, check=True, capture_output=True, text=True, encoding='utf-8')
+        print("ffmpeg conversion successful.")
+        # print(f"ffmpeg stderr: {result.stderr}") # Optional debug
+        return True
+    except FileNotFoundError:
+        print("错误: ffmpeg 未找到，无法转换音频。请确保已在 Docker 环境中安装 ffmpeg。")
+        return False
+    except subprocess.CalledProcessError as e:
+        print(f"错误: ffmpeg 执行失败 (返回码 {e.returncode})。")
+        print(f"ffmpeg stderr: {e.stderr}")
+        return False
+    except Exception as e:
+        print(f"转换音频时发生未知错误: {e}")
+        return False
+
+@app.post("/evaluate/", response_model=VisqolResponse)
+async def evaluate_audio(
+    reference: UploadFile = File(..., description="参考音频文件"),
+    degraded: UploadFile = File(..., description="待评估音频文件"),
+    mode: str = "audio" # 'audio' 或 'speech'
+):
+    """
+    使用 ViSQOL 评估两个音频文件之间的感知相似度。
+    返回预测的平均意见得分 (MOS-LQO)。
+    """
+    if visqol_lib_py is None:
+         raise HTTPException(status_code=500, detail="ViSQOL 库未成功加载。")
+
+    if mode not in ["audio", "speech"]:
+        raise HTTPException(status_code=400, detail="模式参数 'mode' 必须是 'audio' 或 'speech'")
+
+    temp_dir = tempfile.mkdtemp()
+    # Save with original extension first to help ffmpeg identify format
+    ref_temp_orig = os.path.join(temp_dir, f"ref_{reference.filename}") 
+    deg_temp_orig = os.path.join(temp_dir, f"deg_{degraded.filename}")
+    # Define final WAV paths
+    ref_path_wav = os.path.join(temp_dir, "reference.wav")
+    deg_path_wav = os.path.join(temp_dir, "degraded.wav")
+    
+    mos = -1.0
+    vnsim_val = None # 初始化 vnsim
+    fvnsim_val = None # 初始化 fvnsim
+    status_msg = "处理失败"
+    error_msg = None
+
+    try:
+        # 1. 保存原始上传文件
+        ref_content = await reference.read()
+        with open(ref_temp_orig, "wb") as f: f.write(ref_content)
+        deg_content = await degraded.read()
+        with open(deg_temp_orig, "wb") as f: f.write(deg_content)
+        await reference.close()
+        await degraded.close()
+
+        # 2. 确定目标采样率并转换/重采样文件
+        target_sr = 48000 if mode == 'audio' else 16000
+        print(f"目标采样率: {target_sr} Hz for mode '{mode}'")
+        
+        conv_ref_ok = convert_and_resample_audio(ref_temp_orig, ref_path_wav, target_sr)
+        conv_deg_ok = convert_and_resample_audio(deg_temp_orig, deg_path_wav, target_sr)
+        
+        if not (conv_ref_ok and conv_deg_ok):
+            raise HTTPException(status_code=500, detail="使用 ffmpeg 转换或重采样音频文件失败。")
+
+        # 3. 验证转换后的 WAV 文件 (可选)
+        try:
+             ref_info = sf.info(ref_path_wav)
+             deg_info = sf.info(deg_path_wav)
+             if ref_info.samplerate != target_sr or deg_info.samplerate != target_sr:
+                 print(f"警告：ffmpeg 转换后的采样率 ({ref_info.samplerate}/{deg_info.samplerate}) 与目标 ({target_sr}) 不符，可能影响 ViSQOL 结果。")
+        except Exception as audio_e:
+             # 如果 sf.info 失败，可能是 ffmpeg 转换有问题
+             raise HTTPException(status_code=400, detail=f"无法读取转换后的 WAV 文件: {audio_e}")
+
+        # 4. 加载转换/重采样后的音频数据
+        try:
+            print(f"从 WAV 加载音频数据: {ref_path_wav}, {deg_path_wav}")
+            # 确保读取为 float64 类型 (对应 C++ double)
+            ref_data, sr_ref = sf.read(ref_path_wav, dtype='float64')
+            deg_data, sr_deg = sf.read(deg_path_wav, dtype='float64')
+            # 确认采样率是否符合预期 (理论上 ffmpeg 已经处理)
+            if sr_ref != target_sr or sr_deg != target_sr:
+                print(f"警告：读取的 WAV 文件采样率 ({sr_ref}/{sr_deg}) 与目标 ({target_sr}) 不符。")
+                # 可以选择在这里停止或继续
+            print("音频数据加载成功。")
+        except Exception as read_e:
+            raise HTTPException(status_code=500, detail=f"读取转换后的 WAV 文件时出错: {read_e}")
+
+        # 5. 初始化 ViSQOL 配置 (修正模型选择逻辑)
+        config = visqol_config_pb2.VisqolConfig()
+        config.audio.sample_rate = target_sr # 使用目标采样率
+        
+        # 修正模型选择：根据官方示例调整
+        if mode == "speech":
+            config.options.use_speech_scoring = True
+            # Speech mode uses the TFLite model according to official example
+            model_file_to_use = AUDIO_MODEL_PATH # .tflite model
+        else: # audio mode
+            config.options.use_speech_scoring = False
+            # Audio mode uses the SVR model according to official example
+            model_file_to_use = SPEECH_MODEL_PATH # .txt model (libsvm)
+            
+        config.options.svr_model_path = os.path.abspath(model_file_to_use)
+        print(f"使用模型: {model_file_to_use} for mode '{mode}'")
+
+        # 6. 创建 API 实例并运行评估 (传递数据而不是路径)
+        api = visqol_lib_py.VisqolApi()
+        api.Create(config) # 传递对象
+        # 传递加载的 NumPy 数组
+        similarity_result_msg = api.Measure(ref_data, deg_data) # <--- 修改此处
+
+        # 7. 处理结果 (逻辑保持不变，增加提取 vnsim 和 fvnsim)
+        if similarity_result_msg and hasattr(similarity_result_msg, 'moslqo'):
+            mos = similarity_result_msg.moslqo
+            status_msg = "处理成功"
+            print(f"ViSQOL 评估完成: MOS-LQO = {mos}")
+            # 尝试提取 vnsim
+            if hasattr(similarity_result_msg, 'vnsim'):
+                 vnsim_val = similarity_result_msg.vnsim
+                 print(f"VNSIM = {vnsim_val}")
+            else:
+                 print("ViSQOL 结果中未找到 vnsim 字段。")
+            # 尝试提取 fvnsim (需要转换为 Python 列表)
+            if hasattr(similarity_result_msg, 'fvnsim') and similarity_result_msg.fvnsim:
+                 fvnsim_val = list(similarity_result_msg.fvnsim) # 转换为列表
+                 print(f"FVNSIM (第一个元素): {fvnsim_val[0] if fvnsim_val else 'N/A'}") # 打印部分信息
+            else:
+                 print("ViSQOL 结果中未找到 fvnsim 字段或为空。")
+        else:
+            error_msg = "ViSQOL 未返回有效的 MOS-LQO 结果。"
+            print(f"错误: {error_msg}")
+
+    except ImportError as e:
+         status_msg = "导入错误"
+         error_msg = f"无法导入 ViSQOL 库或依赖: {e}"
+         print(f"错误: {error_msg}")
+    except FileNotFoundError as e:
+         status_msg = "文件未找到错误"
+         error_msg = f"必需文件丢失: {e}"
+         print(f"错误: {error_msg}")
+    except HTTPException as e: # 捕获我们自己抛出的 HTTP 异常
+        status_msg = "请求错误"
+        error_msg = str(e.detail)
+        print(f"错误: {error_msg}")
+    except Exception as e:
+        status_msg = "运行时错误"
+        error_msg = f"处理过程中发生错误: {type(e).__name__} - {e}"
+        print(f"错误: {error_msg}")
+        # 可以在这里添加更详细的堆栈跟踪日志，如果需要
+        # import traceback
+        # print(traceback.format_exc())
+    finally:
+        if os.path.exists(temp_dir):
+            shutil.rmtree(temp_dir)
+
+    return VisqolResponse(
+        reference_filename=reference.filename,
+        degraded_filename=degraded.filename,
+        mode=mode,
+        moslqo=mos,
+        vnsim=vnsim_val, # 添加 vnsim 到响应
+        fvnsim=fvnsim_val, # 添加 fvnsim 到响应
+        status=status_msg,
+        error_message=error_msg
+    )
+
+@app.get("/", include_in_schema=False)
+async def root():
+    # 提供一个简单的根路径信息
+    return {"message": "欢迎使用 ViSQOL 音频质量评估 API。请使用 POST 方法访问 /evaluate/ 端点。"}
+
+# 添加健康检查端点
+@app.get("/healthz", status_code=200)
+async def health_check():
+    """Hugging Face Spaces health check endpoint."""
+    # 如果 ViSQOL 库加载失败，也在这里反映出来
+    if visqol_lib_py is None:
+         return {"status": "error", "detail": "ViSQOL library not loaded"}
+    return {"status": "ok"}
+
+# 如果直接运行脚本，用于本地测试 (可选)
+if __name__ == "__main__":
+    import uvicorn
+    print("运行本地测试服务器: http://127.0.0.1:8000")
+    # 注意：本地运行可能需要正确设置 LD_LIBRARY_PATH 或将 .so 文件放在系统可查找的路径
+    uvicorn.run(app, host="127.0.0.1", port=8000) 

build/visqol/model/lattice_tcditugenmeetpackhref_ls2_nl60_lr12_bs2048_learn.005_ep2400_train1_7_raw.tflite

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4bd031d6c95594ae2cd1d471aa6611bff26b3bbf816528431ff1741bcd798dc6
+size 2233840

build/visqol/pb2/similarity_result_pb2.py

@@ -0,0 +1,45 @@
+# -*- coding: utf-8 -*-
+# Generated by the protocol buffer compiler.  DO NOT EDIT!
+# source: similarity_result.proto
+"""Generated protocol buffer code."""
+from google.protobuf import descriptor as _descriptor
+from google.protobuf import descriptor_pool as _descriptor_pool
+from google.protobuf import message as _message
+from google.protobuf import reflection as _reflection
+from google.protobuf import symbol_database as _symbol_database
+# @@protoc_insertion_point(imports)
+
+_sym_db = _symbol_database.Default()
+
+
+
+
+DESCRIPTOR = _descriptor_pool.Default().AddSerializedFile(b'\n\x17similarity_result.proto\x12\x06Visqol\"\xe4\x03\n\x13SimilarityResultMsg\x12\x0e\n\x06moslqo\x18\x01 \x01(\x01\x12\r\n\x05vnsim\x18\x02 \x01(\x01\x12\x0e\n\x06\x66vnsim\x18\x03 \x03(\x01\x12\x10\n\x08\x66vnsim10\x18\x0b \x03(\x01\x12\x10\n\x08\x66stdnsim\x18\x08 \x03(\x01\x12\x13\n\x0b\x66vdegenergy\x18\t \x03(\x01\x12\x19\n\x11\x63\x65nter_freq_bands\x18\x04 \x03(\x01\x12\x42\n\npatch_sims\x18\x05 \x03(\x0b\x32..Visqol.SimilarityResultMsg.PatchSimilarityMsg\x12\x1a\n\x12reference_filepath\x18\x06 \x01(\t\x12\x19\n\x11\x64\x65graded_filepath\x18\x07 \x01(\t\x12\x17\n\x0f\x61lignment_lag_s\x18\n \x01(\x01\x1a\xb5\x01\n\x12PatchSimilarityMsg\x12\x12\n\nsimilarity\x18\x01 \x01(\x01\x12\x17\n\x0f\x66req_band_means\x18\x02 \x03(\x01\x12\x1c\n\x14ref_patch_start_time\x18\x03 \x01(\x01\x12\x1a\n\x12ref_patch_end_time\x18\x04 \x01(\x01\x12\x1c\n\x14\x64\x65g_patch_start_time\x18\x05 \x01(\x01\x12\x1a\n\x12\x64\x65g_patch_end_time\x18\x06 \x01(\x01\x62\x06proto3')
+
+
+
+_SIMILARITYRESULTMSG = DESCRIPTOR.message_types_by_name['SimilarityResultMsg']
+_SIMILARITYRESULTMSG_PATCHSIMILARITYMSG = _SIMILARITYRESULTMSG.nested_types_by_name['PatchSimilarityMsg']
+SimilarityResultMsg = _reflection.GeneratedProtocolMessageType('SimilarityResultMsg', (_message.Message,), {
+
+  'PatchSimilarityMsg' : _reflection.GeneratedProtocolMessageType('PatchSimilarityMsg', (_message.Message,), {
+    'DESCRIPTOR' : _SIMILARITYRESULTMSG_PATCHSIMILARITYMSG,
+    '__module__' : 'similarity_result_pb2'
+    # @@protoc_insertion_point(class_scope:Visqol.SimilarityResultMsg.PatchSimilarityMsg)
+    })
+  ,
+  'DESCRIPTOR' : _SIMILARITYRESULTMSG,
+  '__module__' : 'similarity_result_pb2'
+  # @@protoc_insertion_point(class_scope:Visqol.SimilarityResultMsg)
+  })
+_sym_db.RegisterMessage(SimilarityResultMsg)
+_sym_db.RegisterMessage(SimilarityResultMsg.PatchSimilarityMsg)
+
+if _descriptor._USE_C_DESCRIPTORS == False:
+
+  DESCRIPTOR._options = None
+  _SIMILARITYRESULTMSG._serialized_start=36
+  _SIMILARITYRESULTMSG._serialized_end=520
+  _SIMILARITYRESULTMSG_PATCHSIMILARITYMSG._serialized_start=339
+  _SIMILARITYRESULTMSG_PATCHSIMILARITYMSG._serialized_end=520
+# @@protoc_insertion_point(module_scope)

build/visqol/pb2/visqol_config_pb2.py

@@ -0,0 +1,56 @@
+# -*- coding: utf-8 -*-
+# Generated by the protocol buffer compiler.  DO NOT EDIT!
+# source: visqol_config.proto
+"""Generated protocol buffer code."""
+from google.protobuf import descriptor as _descriptor
+from google.protobuf import descriptor_pool as _descriptor_pool
+from google.protobuf import message as _message
+from google.protobuf import reflection as _reflection
+from google.protobuf import symbol_database as _symbol_database
+# @@protoc_insertion_point(imports)
+
+_sym_db = _symbol_database.Default()
+
+
+
+
+DESCRIPTOR = _descriptor_pool.Default().AddSerializedFile(b'\n\x13visqol_config.proto\x12\x06Visqol\"\xa9\x03\n\x0cVisqolConfig\x12\x33\n\x05\x61udio\x18\x01 \x01(\x0b\x32$.Visqol.VisqolConfig.VisqolAudioInfo\x12\x33\n\x07options\x18\x02 \x01(\x0b\x32\".Visqol.VisqolConfig.VisqolOptions\x1a&\n\x0fVisqolAudioInfo\x12\x13\n\x0bsample_rate\x18\x01 \x01(\x05\x1a\x86\x02\n\rVisqolOptions\x12\x18\n\x10output_mos_score\x18\x01 \x01(\x08\x12\x16\n\x0esvr_model_path\x18\x02 \x01(\t\x12\x1a\n\x12use_speech_scoring\x18\x03 \x01(\x08\x12\x1d\n\x15\x64\x65tect_voice_activity\x18\x04 \x01(\x08\x12&\n\x1e\x61llow_unsupported_sample_rates\x18\x05 \x01(\x08\x12\'\n\x1fuse_unscaled_speech_mos_mapping\x18\x06 \x01(\x08\x12\x1c\n\x14search_window_radius\x18\x07 \x01(\x05\x12\x19\n\x11use_lattice_model\x18\x08 \x01(\x08\x62\x06proto3')
+
+
+
+_VISQOLCONFIG = DESCRIPTOR.message_types_by_name['VisqolConfig']
+_VISQOLCONFIG_VISQOLAUDIOINFO = _VISQOLCONFIG.nested_types_by_name['VisqolAudioInfo']
+_VISQOLCONFIG_VISQOLOPTIONS = _VISQOLCONFIG.nested_types_by_name['VisqolOptions']
+VisqolConfig = _reflection.GeneratedProtocolMessageType('VisqolConfig', (_message.Message,), {
+
+  'VisqolAudioInfo' : _reflection.GeneratedProtocolMessageType('VisqolAudioInfo', (_message.Message,), {
+    'DESCRIPTOR' : _VISQOLCONFIG_VISQOLAUDIOINFO,
+    '__module__' : 'visqol_config_pb2'
+    # @@protoc_insertion_point(class_scope:Visqol.VisqolConfig.VisqolAudioInfo)
+    })
+  ,
+
+  'VisqolOptions' : _reflection.GeneratedProtocolMessageType('VisqolOptions', (_message.Message,), {
+    'DESCRIPTOR' : _VISQOLCONFIG_VISQOLOPTIONS,
+    '__module__' : 'visqol_config_pb2'
+    # @@protoc_insertion_point(class_scope:Visqol.VisqolConfig.VisqolOptions)
+    })
+  ,
+  'DESCRIPTOR' : _VISQOLCONFIG,
+  '__module__' : 'visqol_config_pb2'
+  # @@protoc_insertion_point(class_scope:Visqol.VisqolConfig)
+  })
+_sym_db.RegisterMessage(VisqolConfig)
+_sym_db.RegisterMessage(VisqolConfig.VisqolAudioInfo)
+_sym_db.RegisterMessage(VisqolConfig.VisqolOptions)
+
+if _descriptor._USE_C_DESCRIPTORS == False:
+
+  DESCRIPTOR._options = None
+  _VISQOLCONFIG._serialized_start=32
+  _VISQOLCONFIG._serialized_end=457
+  _VISQOLCONFIG_VISQOLAUDIOINFO._serialized_start=154
+  _VISQOLCONFIG_VISQOLAUDIOINFO._serialized_end=192
+  _VISQOLCONFIG_VISQOLOPTIONS._serialized_start=195
+  _VISQOLCONFIG_VISQOLOPTIONS._serialized_end=457
+# @@protoc_insertion_point(module_scope)

build/visqol/visqol_lib_py.so

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a0e70dc0a46db8549b183819ef543d22d665952ebd999d75506c7969d46a05ba
+size 10110248

requirements.txt

@@ -0,0 +1,8 @@
+fastapi
+uvicorn[standard]
+python-multipart
+pydantic
+numpy
+soundfile
+protobuf~=3.20.0
+librosa