Upload scripts/rvc_train_colab.ipynb with huggingface_hub

scripts/rvc_train_colab.ipynb

@@ -1,284 +1,244 @@
 {
- "nbformat": 4,
- "nbformat_minor": 0,
- "metadata": {
-  "colab": {
-   "provenance": [],
-   "gpuType": "T4"
+  "nbformat": 4,
+  "nbformat_minor": 0,
+  "metadata": {
+    "colab": {
+      "provenance": [],
+      "gpuType": "T4"
+    },
+    "kernelspec": {
+      "name": "python3",
+      "display_name": "Python 3"
+    },
+    "language_info": {
+      "name": "python"
+    }
   },
-  "kernelspec": {
-   "name": "python3",
-   "display_name": "Python 3"
-  },
-  "accelerator": "GPU"
- },
- "cells": [
-  {
-   "cell_type": "markdown",
-   "metadata": {},
-   "source": [
-    "# NumberBlocks One - RVC Voice Cloning Training\n",
-    "\n",
-    "一键训练脚本：自动下载 2607 个训练片段，训练 RVC 模型，上传模型到 HuggingFace。\n",
-    "\n",
-    "**使用方法**: Runtime → Change runtime type → T4 GPU → Run All\n",
-    "\n",
-    "预计训练时间：~30-60分钟 (T4 GPU, 10000 steps)"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "#@title 1. 环境检查\n",
-    "import subprocess\n",
-    "result = subprocess.run(['nvidia-smi'], capture_output=True, text=True)\n",
-    "if result.returncode != 0:\n",
-    "    print('⚠️ No GPU detected! Go to Runtime → Change runtime type → T4 GPU')\n",
-    "else:\n",
-    "    print('✅ GPU detected:')\n",
-    "    print(result.stdout[:500])"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "#@title 2. 安装 RVC 和依赖 (~5 min)\n",
-    "%cd /content\n",
-    "\n",
-    "# Install RVC\n",
-    "!git clone https://github.com/RVC-Project/Retrieval-based-Voice-Conversion.git RVC\n",
-    "%cd RVC\n",
-    "!pip install -r requirements.txt -q\n",
-    "!pip install praat-parselmouth -q\n",
-    "\n",
-    "# Install ffmpeg\n",
-    "!apt-get install -y ffmpeg -qq\n",
-    "\n",
-    "print('✅ RVC installed')"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "#@title 3. 下载训练数据 (2607 segments from HuggingFace)\n",
-    "import os\n",
-    "from huggingface_hub import HfApi, hf_hub_download\n",
-    "\n",
-    "DATASET_REPO = 'ayf3/numberblocks-one-voice-dataset'\n",
-    "WORK_DIR = '/content/RVC/dataset/numberblocks_one'\n",
-    "os.makedirs(WORK_DIR, exist_ok=True)\n",
-    "\n",
-    "api = HfApi()\n",
-    "files = api.list_repo_files(DATASET_REPO, repo_type='dataset')\n",
-    "segments = [f for f in files if f.startswith('segments/') and f.endswith('.wav')]\n",
-    "\n",
-    "print(f'Downloading {len(segments)} training segments...')\n",
-    "\n",
-    "# Download in parallel using huggingface_hub\n",
-    "from concurrent.futures import ThreadPoolExecutor, as_completed\n",
-    "\n",
-    "def download_seg(seg_path):\n",
-    "    local = hf_hub_download(DATASET_REPO, seg_path, repo_type='dataset',\n",
-    "                           local_dir='/content/rvc_data')\n",
-    "    return local\n",
-    "\n",
-    "# Download all segments to a temp dir, then we'll point RVC to it\n",
-    "import shutil\n",
-    "TEMP_DIR = '/content/rvc_data/segments'\n",
-    "os.makedirs(TEMP_DIR, exist_ok=True)\n",
-    "\n",
-    "downloaded = 0\n",
-    "with ThreadPoolExecutor(max_workers=8) as executor:\n",
-    "    futures = {executor.submit(download_seg, s): s for s in segments}\n",
-    "    for future in as_completed(futures):\n",
-    "        downloaded += 1\n",
-    "        if downloaded % 200 == 0:\n",
-    "            print(f'  Downloaded {downloaded}/{len(segments)}...')\n",
-    "\n",
-    "print(f'✅ Downloaded {downloaded} segments')\n",
-    "\n",
-    "# Copy to RVC dataset dir with clean names\n",
-    "for f in os.listdir(TEMP_DIR):\n",
-    "    if f.endswith('.wav'):\n",
-    "        src = os.path.join(TEMP_DIR, f)\n",
-    "        dst = os.path.join(WORK_DIR, f)\n",
-    "        if not os.path.exists(dst):\n",
-    "            shutil.copy2(src, dst)\n",
-    "\n",
-    "segment_files = [f for f in os.listdir(WORK_DIR) if f.endswith('.wav')]\n",
-    "print(f'✅ {len(segment_files)} segments ready in {WORK_DIR}')\n",
-    "\n",
-    "# Calculate total audio duration\n",
-    "total_dur = 0\n",
-    "for f in segment_files[:10]:  # Sample to estimate\n",
-    "    result = subprocess.run(\n",
-    "        ['ffprobe', '-v', 'quiet', '-show_entries', 'format=duration',\n",
-    "         '-of', 'csv=p=0', os.path.join(WORK_DIR, f)],\n",
-    "        capture_output=True, text=True\n",
-    "    )\n",
-    "    if result.stdout.strip():\n",
-    "        total_dur += float(result.stdout.strip())\n",
-    "avg_dur = total_dur / min(10, len(segment_files))\n",
-    "est_total = avg_dur * len(segment_files) / 60\n",
-    "print(f'📊 Estimated total audio: {est_total:.1f} minutes')"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "#@title 4. 特征提取 (~10 min)\n",
-    "%cd /content/RVC\n",
-    "\n",
-    "MODEL_NAME = 'numberblocks_one'\n",
-    "DATASET_PATH = '/content/RVC/dataset/numberblocks_one'\n",
-    "EXP_DIR = f'/content/RVC/logs/{MODEL_NAME}'\n",
-    "os.makedirs(EXP_DIR, exist_ok=True)\n",
-    "\n",
-    "# Run feature extraction\n",
-    "!python infer/modules/train/extract/extract_f0_print.py \\\n",
-    "    {EXP_DIR} \\\n",
-    "    {DATASET_PATH} \\\n",
-    "    harvest \\\n",
-    "    0 \\\n",
-    "    8\n",
-    "\n",
-    "print('F0 extraction done')\n",
-    "\n",
-    "# Extract features\n",
-    "!python infer/modules/train/extract/extract_feature_print.py \\\n",
-    "    {EXP_DIR} \\\n",
-    "    v2 \\\n",
-    "    0 \\\n",
-    "    8\n",
-    "\n",
-    "print('✅ Feature extraction complete')"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "#@title 5. 训练 RVC 模型 (~20-40 min)\n",
-    "import os\n",
-    "\n",
-    "MODEL_NAME = 'numberblocks_one'\n",
-    "EXP_DIR = f'/content/RVC/logs/{MODEL_NAME}'\n",
-    "\n",
-    "# Check how many features were extracted\n",
-    "f0_files = [f for f in os.listdir(EXP_DIR) if f.endswith('.f0.npy') or f.endswith('.f0')]\n",
-    "feat_files = [f for f in os.listdir(EXP_DIR) if 'feature' in f.lower()]\n",
-    "print(f'F0 files: {len(f0_files)}, Feature files found')\n",
-    "\n",
-    "# List all files in EXP_DIR\n",
-    "all_files = os.listdir(EXP_DIR)\n",
-    "print(f'Total files in {EXP_DIR}: {len(all_files)}')\n",
-    "for f in sorted(all_files)[:20]:\n",
-    "    print(f'  {f}')\n",
-    "\n",
-    "# Train with config\n",
-    "SAVE_EVERY = 1000  #@param {type:\"number\"}\n",
-    "TOTAL_STEPS = 10000  #@param {type:\"number\"}\n",
-    "BATCH_SIZE = 8  #@param {type:\"number\"}\n",
-    "\n",
-    "%cd /content/RVC\n",
-    "\n",
-    "!python infer/modules/train/train.py \\\n",
-    "    -e {MODEL_NAME} \\\n",
-    "    -s {TOTAL_STEPS} \\\n",
-    "    -b {BATCH_SIZE} \\\n",
-    "    -sr 40k \\\n",
-    "    -v v2 \\\n",
-    "    -g 0 \\\n",
-    "    -se {SAVE_EVERY}\n",
-    "\n",
-    "print('✅ Training complete!')"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "#@title 6. 上传模型到 HuggingFace\n",
-    "import os\n",
-    "import glob\n",
-    "from huggingface_hub import HfApi\n",
-    "\n",
-    "MODEL_NAME = 'numberblocks_one'\n",
-    "DATASET_REPO = 'ayf3/numberblocks-one-voice-dataset'\n",
-    "\n",
-    "# Find the trained model\n",
-    "model_dir = f'/content/RVC/weights'\n",
-    "pth_files = glob.glob(f'{model_dir}/**/*{MODEL_NAME}*.pth', recursive=True)\n",
-    "\n",
-    "if not pth_files:\n",
-    "    # Check in logs dir\n",
-    "    pth_files = glob.glob(f'/content/RVC/logs/{MODEL_NAME}/*.pth')\n",
-    "    if not pth_files:\n",
-    "        pth_files = glob.glob(f'/content/RVC/logs/{MODEL_NAME}/**/*.pth', recursive=True)\n",
-    "\n",
-    "if pth_files:\n",
-    "    print(f'Found model files:')\n",
-    "    for p in pth_files:\n",
-    "        size_mb = os.path.getsize(p) / 1024 / 1024\n",
-    "        print(f'  {p} ({size_mb:.1f} MB)')\n",
-    "    \n",
-    "    # Upload the best model (largest .pth)\n",
-    "    best = max(pth_files, key=lambda x: os.path.getsize(x))\n",
-    "    print(f'\\nUploading {best} to HuggingFace...')\n",
-    "    \n",
-    "    api = HfApi()\n",
-    "    api.upload_file(\n",
-    "        path_or_fileobj=best,\n",
-    "        path_in_repo=f'models/{os.path.basename(best)}',\n",
-    "        repo_id=DATASET_REPO,\n",
-    "        repo_type='dataset',\n",
-    "    )\n",
-    "    print(f'✅ Model uploaded to {DATASET_REPO}/models/{os.path.basename(best)}')\n",
-    "    \n",
-    "    # Also upload feature index if exists\n",
-    "    index_files = glob.glob(f'/content/RVC/logs/{MODEL_NAME}/**/*.index', recursive=True)\n",
-    "    for idx_file in index_files:\n",
-    "        api.upload_file(\n",
-    "            path_or_fileobj=idx_file,\n",
-    "            path_in_repo=f'models/{os.path.basename(idx_file)}',\n",
-    "            repo_id=DATASET_REPO,\n",
-    "            repo_type='dataset',\n",
-    "        )\n",
-    "        print(f'✅ Index uploaded: {os.path.basename(idx_file)}')\n",
-    "else:\n",
-    "    print('❌ No .pth model files found. Training may have failed.')\n",
-    "    # List all files to debug\n",
-    "    for root, dirs, files_in_dir in os.walk(f'/content/RVC/logs/{MODEL_NAME}'):\n",
-    "        for f in files_in_dir:\n",
-    "            print(f'  {os.path.join(root, f)}')"
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "metadata": {},
-   "source": [
-    "## 🎉 完成！\n",
-    "\n",
-    "训练好的模型已上传到 HuggingFace dataset 的 `models/` 目录。\n",
-    "\n",
-    "**下一步**: 使用训练好的模型进行 voice conversion 推理。"
-   ]
-  }
- ]
+  "cells": [
+    {
+      "cell_type": "markdown",
+      "metadata": {},
+      "source": [
+        "# NumberBlocks One - RVC v2 Voice Cloning Training\n",
+        "\n",
+        "\u4e00\u952e\u8bad\u7ec3\uff1a\u81ea\u52a8\u4e0b\u8f7d\u6570\u636e \u2192 \u7279\u5f81\u63d0\u53d6 \u2192 \u8bad\u7ec3 RVC v2 \u6a21\u578b \u2192 \u4e0a\u4f20\u5230 HuggingFace\u3002\n",
+        "\n",
+        "**\u4f7f\u7528**: Runtime \u2192 Change runtime type \u2192 **T4 GPU** \u2192 Run All\n",
+        "\n",
+        "\u9884\u8ba1: ~30-60min (T4, 5000 steps on top500+augmented)\n",
+        "\n",
+        "\u6570\u636e\u6e90: 500 \u9ad8\u8d28\u91cf segments + 1500 augmented = **2000 \u8bad\u7ec3\u6837\u672c**\n"
+      ]
+    },
+    {
+      "cell_type": "code",
+      "metadata": {},
+      "source": [
+        "#@title 1. \u73af\u5883\u68c0\u67e5\n",
+        "import subprocess, os\n",
+        "result = subprocess.run(['nvidia-smi'], capture_output=True, text=True)\n",
+        "if result.returncode != 0:\n",
+        "    print('\u26a0\ufe0f No GPU! Runtime \u2192 Change runtime type \u2192 T4 GPU')\n",
+        "    raise RuntimeError('GPU required')\n",
+        "else:\n",
+        "    print('\u2705 GPU detected')\n",
+        "    lines = result.stdout.split('\\n')\n",
+        "    print(lines[8] if len(lines) > 8 else result.stdout[:300])\n"
+      ],
+      "outputs": [],
+      "execution_count": null
+    },
+    {
+      "cell_type": "code",
+      "metadata": {},
+      "source": [
+        "#@title 2. \u5b89\u88c5 RVC \u548c\u4f9d\u8d56 (~5 min)\n",
+        "import os, subprocess, sys\n",
+        "\n",
+        "%cd /content\n",
+        "\n",
+        "if not os.path.exists('/content/RVC'):\n",
+        "    !git clone --depth 1 https://github.com/RVC-Project/Retrieval-based-Voice-Conversion.git RVC\n",
+        "\n",
+        "%cd RVC\n",
+        "subprocess.run([sys.executable, '-m', 'pip', 'install', '-r', 'requirements.txt', '-q'], check=False)\n",
+        "subprocess.run([sys.executable, '-m', 'pip', 'install', 'praat-parselmouth', '-q'], check=False)\n",
+        "subprocess.run(['apt-get', 'install', '-y', 'ffmpeg', '-qq'], check=False)\n",
+        "\n",
+        "print('\u2705 RVC installed')\n"
+      ],
+      "outputs": [],
+      "execution_count": null
+    },
+    {
+      "cell_type": "code",
+      "metadata": {},
+      "source": [
+        "#@title 3. \u4e0b\u8f7d\u8bad\u7ec3\u6570\u636e (top500 + augmented = ~2000 files)\n",
+        "import os, shutil, concurrent.futures\n",
+        "from huggingface_hub import HfApi, hf_hub_download\n",
+        "\n",
+        "DATASET_REPO = 'ayf3/numberblocks-one-voice-dataset'\n",
+        "WORK_DIR = '/content/RVC/dataset/numberblocks_one'\n",
+        "os.makedirs(WORK_DIR, exist_ok=True)\n",
+        "\n",
+        "api = HfApi()\n",
+        "all_files = list(api.list_repo_files(DATASET_REPO, repo_type='dataset'))\n",
+        "\n",
+        "train_files = [f for f in all_files\n",
+        "    if (f.startswith('data/train_top500/') or f.startswith('data/train_augmented/'))\n",
+        "    and f.endswith('.wav')]\n",
+        "\n",
+        "print(f'Total training files to download: {len(train_files)}')\n",
+        "\n",
+        "def download_one(f):\n",
+        "    try:\n",
+        "        local = hf_hub_download(DATASET_REPO, f, repo_type='dataset')\n",
+        "        basename = os.path.basename(f)\n",
+        "        shutil.copy2(local, os.path.join(WORK_DIR, basename))\n",
+        "        return True\n",
+        "    except:\n",
+        "        return False\n",
+        "\n",
+        "with concurrent.futures.ThreadPoolExecutor(max_workers=8) as executor:\n",
+        "    results = list(executor.map(download_one, train_files))\n",
+        "\n",
+        "success = sum(1 for r in results if r)\n",
+        "existing = [f for f in os.listdir(WORK_DIR) if f.endswith('.wav')]\n",
+        "print(f'\u2705 Downloaded: {success}/{len(train_files)}, Files in dataset dir: {len(existing)}')\n"
+      ],
+      "outputs": [],
+      "execution_count": null
+    },
+    {
+      "cell_type": "code",
+      "metadata": {},
+      "source": [
+        "#@title 4. \u7279\u5f81\u63d0\u53d6 (~10-15 min)\n",
+        "import os, subprocess, sys\n",
+        "\n",
+        "%cd /content/RVC\n",
+        "\n",
+        "MODEL_NAME = 'numberblocks_one'\n",
+        "DATASET_PATH = '/content/RVC/dataset/numberblocks_one'\n",
+        "EXP_DIR = f'/content/RVC/logs/{MODEL_NAME}'\n",
+        "os.makedirs(EXP_DIR, exist_ok=True)\n",
+        "\n",
+        "wav_files = [f for f in os.listdir(DATASET_PATH) if f.endswith('.wav')]\n",
+        "print(f'Training WAV files: {len(wav_files)}')\n",
+        "\n",
+        "# F0 extraction\n",
+        "print('\\n--- Extracting F0 (harvest) ---')\n",
+        "r = subprocess.run([sys.executable, 'infer/modules/train/extract/extract_f0_print.py',\n",
+        "    EXP_DIR, DATASET_PATH, 'harvest', '0', '8'], capture_output=True, text=True, timeout=1800)\n",
+        "print(r.stdout[-500:] if r.stdout else '')\n",
+        "print(f'F0 return code: {r.returncode}')\n",
+        "\n",
+        "# Feature extraction\n",
+        "print('\\n--- Extracting features (v2) ---')\n",
+        "r = subprocess.run([sys.executable, 'infer/modules/train/extract/extract_feature_print.py',\n",
+        "    EXP_DIR, 'v2', DATASET_PATH, '0', '8', 'fp32'], capture_output=True, text=True, timeout=1800)\n",
+        "print(r.stdout[-500:] if r.stdout else '')\n",
+        "print(f'Feature return code: {r.returncode}')\n",
+        "\n",
+        "exp_files = os.listdir(EXP_DIR)\n",
+        "print(f'\\n\u2705 Files in experiment dir: {len(exp_files)}')\n"
+      ],
+      "outputs": [],
+      "execution_count": null
+    },
+    {
+      "cell_type": "code",
+      "metadata": {},
+      "source": [
+        "#@title 5. \u8bad\u7ec3 RVC v2 \u6a21\u578b (~20-40 min)\n",
+        "import os, subprocess, sys, glob\n",
+        "\n",
+        "%cd /content/RVC\n",
+        "\n",
+        "MODEL_NAME = 'numberblocks_one'\n",
+        "EXP_DIR = f'/content/RVC/logs/{MODEL_NAME}'\n",
+        "\n",
+        "cmd = [sys.executable, 'infer/modules/train/train.py',\n",
+        "    '-e', EXP_DIR, '-n', MODEL_NAME,\n",
+        "    '-sr', '40k', '-bs', '8', '-g', '0',\n",
+        "    '-te', '5000', '-se', '500',\n",
+        "    '-v', 'v2', '-l', '1']\n",
+        "\n",
+        "print(f'Training: {\" \".join(cmd)}')\n",
+        "r = subprocess.run(cmd, capture_output=True, text=True, timeout=7200)\n",
+        "print(r.stdout[-1000:] if r.stdout else '')\n",
+        "if r.returncode != 0:\n",
+        "    print('STDERR:', r.stderr[-500:])\n",
+        "\n",
+        "pth_files = glob.glob(f'{EXP_DIR}/**/*.pth', recursive=True)\n",
+        "print(f'\\n\u2705 PTH files: {len(pth_files)}')\n",
+        "for p in pth_files:\n",
+        "    print(f'  {p} ({os.path.getsize(p)/1024/1024:.1f} MB)')\n"
+      ],
+      "outputs": [],
+      "execution_count": null
+    },
+    {
+      "cell_type": "code",
+      "metadata": {},
+      "source": [
+        "#@title 6. \u4e0a\u4f20\u6a21\u578b\u5230 HuggingFace\n",
+        "import os, glob, json\n",
+        "from huggingface_hub import HfApi\n",
+        "\n",
+        "DATASET_REPO = 'ayf3/numberblocks-one-voice-dataset'\n",
+        "api = HfApi()\n",
+        "\n",
+        "pth_files = glob.glob('/content/RVC/**/*.pth', recursive=True)\n",
+        "print(f'Found {len(pth_files)} .pth files')\n",
+        "\n",
+        "if not pth_files:\n",
+        "    print('\u26a0\ufe0f No model files found!')\n",
+        "else:\n",
+        "    for pth_path in pth_files:\n",
+        "        fname = os.path.basename(pth_path)\n",
+        "        size_mb = os.path.getsize(pth_path) / 1024 / 1024\n",
+        "        print(f'Uploading {fname} ({size_mb:.1f} MB)...')\n",
+        "        api.upload_file(\n",
+        "            path_or_fileobj=pth_path,\n",
+        "            path_in_repo=f'models/{fname}',\n",
+        "            repo_id=DATASET_REPO,\n",
+        "            repo_type='dataset',\n",
+        "        )\n",
+        "        print(f'  \u2705 models/{fname}')\n",
+        "\n",
+        "    # Upload config\n",
+        "    config = {\n",
+        "        'model_name': 'numberblocks_one',\n",
+        "        'version': 'v2',\n",
+        "        'sample_rate': 40000,\n",
+        "        'training_files': 2000,\n",
+        "        'data_source': 'top500 + augmented',\n",
+        "        'total_steps': 5000,\n",
+        "        'batch_size': 8,\n",
+        "    }\n",
+        "    with open('/tmp/train_config.json', 'w') as f:\n",
+        "        json.dump(config, f, indent=2)\n",
+        "    api.upload_file(\n",
+        "        path_or_fileobj='/tmp/train_config.json',\n",
+        "        path_in_repo='models/train_config.json',\n",
+        "        repo_id=DATASET_REPO,\n",
+        "        repo_type='dataset',\n",
+        "    )\n",
+        "    print('\u2705 Config uploaded')\n"
+      ],
+      "outputs": [],
+      "execution_count": null
+    },
+    {
+      "cell_type": "markdown",
+      "metadata": {},
+      "source": [
+        "## \ud83c\udf89 \u5b8c\u6210\uff01\n",
+        "\n",
+        "\u6a21\u578b\u5df2\u4e0a\u4f20\u5230 `models/` \u76ee\u5f55\u3002\u4e0b\u4e00\u6b65\uff1avoice conversion \u63a8\u7406\u3002\n"
+      ]
+    }
+  ]
 }