Add/fix kaggle_fix_dataset.ipynb (Hindi metadata bug fix, speaker IDs, Kaggle fix script)

kaggle_fix_dataset.ipynb

@@ -0,0 +1,175 @@
+{
+ "cells": [
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "# 🔧 ViuAI_TTS_Dataset Fix — Kaggle se (no local work)\n",
+    "\n",
+    "**Problem:** HF repo `ViuAI/ViuAI_TTS_Dataset` mein `audio/` mein 3462 WAVs hain lekin `metadata.csv` mein sirf 2000 rows — 1462 orphan files bina transcript ke padi hain (purane collector bug se).\n",
+    "\n",
+    "**Yeh notebook karega:**\n",
+    "1. HF se dataset download\n",
+    "2. Audit: WAVs vs metadata rows compare\n",
+    "3. Orphan WAVs ko quarantine (delete nahi, safe)\n",
+    "4. Metadata saaf karke wapas HF par upload\n",
+    "\n",
+    "### ⚙️ Pehle ye 2 setting karo (Run se pehle, warna fail hoga):\n",
+    "- **Internet ON:** Right panel → Settings → Internet → ON\n",
+    "- **HF Token:** Add-ons → Secrets → `HF_TOKEN` naam se Write-access token add karo\n"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# Cell 1: Dependencies\n",
+    "!pip install -q datasets huggingface_hub soundfile"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# Cell 2: HF Login (Kaggle Secrets -> env fallback)\n",
+    "import os\n",
+    "token = None\n",
+    "try:\n",
+    "    from kaggle_secrets import UserSecretsClient\n",
+    "    token = UserSecretsClient().get_secret('HF_TOKEN')\n",
+    "except Exception as e:\n",
+    "    print('Secrets unavailable, trying env:', e)\n",
+    "token = token or os.environ.get('HF_TOKEN')\n",
+    "assert token, 'HF_TOKEN nahi mila! Add-ons -> Secrets me HF_TOKEN add karo.'\n",
+    "from huggingface_hub import login, whoami\n",
+    "login(token=token)\n",
+    "print('Logged in as:', whoami(token=token)['name'])"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# Cell 3: Dataset repo download (HF = source of truth)\n",
+    "from huggingface_hub import snapshot_download\n",
+    "REPO = 'ViuAI/ViuAI_TTS_Dataset'\n",
+    "LOCAL = './hf_dataset'\n",
+    "snapshot_download(repo_id=REPO, repo_type='dataset', local_dir=LOCAL)\n",
+    "!ls -la ./hf_dataset && echo '---' && ls ./hf_dataset/audio | wc -l"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# Cell 4: AUDIT (kuch delete nahi hoga, sirf report)\n",
+    "import csv, os, glob\n",
+    "LOCAL = './hf_dataset'\n",
+    "wavs = {os.path.basename(p) for p in glob.glob(os.path.join(LOCAL, 'audio', '*.wav'))}\n",
+    "rows = []\n",
+    "with open(os.path.join(LOCAL, 'metadata.csv'), encoding='utf-8') as f:\n",
+    "    for row in csv.reader(f, delimiter='|'):\n",
+    "        if len(row) >= 4 and row[0] != 'audio_file':\n",
+    "            rows.append(row)\n",
+    "meta_files = {os.path.basename(r[0]) for r in rows}\n",
+    "orphans = sorted(wavs - meta_files)          # disk par hai, metadata me nahi\n",
+    "missing = sorted(meta_files - wavs)          # metadata me hai, disk par nahi\n",
+    "print(f'Disk WAVs      : {len(wavs)}')\n",
+    "print(f'Metadata rows  : {len(rows)}')\n",
+    "print(f'Orphan WAVs    : {len(orphans)} (transcript nahi, training me bekaar)')\n",
+    "print(f'Missing files  : {len(missing)} (metadata row without audio)')\n",
+    "print('Sample orphans :', orphans[:5])\n",
+    "print('Sample missing :', missing[:5])"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# Cell 5: FIX — orphans quarantine + metadata rebuild (local copy par)\n",
+    "# NOTE: Kuch bhi permanent delete nahi hota. Orphans ./quarantine me jate hain\n",
+    "# (Kaggle session ke sath discard). HF se hatane ke liye Cell 6 chalao.\n",
+    "import csv, os, glob, shutil\n",
+    "from collections import Counter\n",
+    "LOCAL = './hf_dataset'\n",
+    "QDIR = './quarantine'\n",
+    "os.makedirs(QDIR, exist_ok=True)\n",
+    "wavs = {os.path.basename(p) for p in glob.glob(os.path.join(LOCAL, 'audio', '*.wav'))}\n",
+    "rows = []\n",
+    "with open(os.path.join(LOCAL, 'metadata.csv'), encoding='utf-8') as f:\n",
+    "    for row in csv.reader(f, delimiter='|'):\n",
+    "        if len(row) >= 4 and row[0] != 'audio_file':\n",
+    "            rows.append(row)\n",
+    "meta_files = {os.path.basename(r[0]) for r in rows}\n",
+    "orphans = sorted(wavs - meta_files)\n",
+    "for fn in orphans:\n",
+    "    shutil.move(os.path.join(LOCAL, 'audio', fn), os.path.join(QDIR, fn))\n",
+    "# Metadata dobara likho: sirf wahi rows jinki audio disk par hai\n",
+    "disk_now = {os.path.basename(p) for p in glob.glob(os.path.join(LOCAL, 'audio', '*.wav'))}\n",
+    "final = [r for r in rows if os.path.basename(r[0]) in disk_now]\n",
+    "dropped = len(rows) - len(final)\n",
+    "with open(os.path.join(LOCAL, 'metadata.csv'), 'w', encoding='utf-8', newline='') as f:\n",
+    "    w = csv.writer(f, delimiter='|')\n",
+    "    w.writerow(['audio_file', 'transcript', 'speaker', 'language'])\n",
+    "    w.writerows(final)\n",
+    "print(f'Quarantined orphans : {len(orphans)}')\n",
+    "print(f'Dropped stale rows  : {dropped}')\n",
+    "print(f'FINAL clips         : {len(final)} (WAVs == rows, 1:1 match)')\n",
+    "print('By language:', dict(Counter(r[3] for r in final)))\n",
+    "print('Speakers:', sorted({r[2] for r in final}))"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# Cell 6: HF par push — pehle orphan delete, phir clean upload\n",
+    "# DANGER ZONE: Cell 5 ke numbers verify karne ke baad hi chalao!\n",
+    "# FINAL clips == disk WAVs hona chahiye (Cell 5 output dekho).\n",
+    "import csv, glob, os\n",
+    "from huggingface_hub import HfApi\n",
+    "REPO = 'ViuAI/ViuAI_TTS_Dataset'\n",
+    "LOCAL = './hf_dataset'\n",
+    "api = HfApi(token=token)\n",
+    "# 1. Remote orphan WAVs delete (batch me, 200 per commit)\n",
+    "remote_files = api.list_repo_files(repo_id=REPO, repo_type='dataset')\n",
+    "remote_wavs = [f for f in remote_files if f.startswith('audio/') and f.endswith('.wav')]\n",
+    "keep = set()\n",
+    "with open(os.path.join(LOCAL, 'metadata.csv'), encoding='utf-8') as f:\n",
+    "    for row in csv.reader(f, delimiter='|'):\n",
+    "        if len(row) >= 1 and row[0] != 'audio_file':\n",
+    "            keep.add(row[0].strip())\n",
+    "to_delete = [f for f in remote_wavs if f not in keep]\n",
+    "print(f'Remote WAVs: {len(remote_wavs)}, keep: {len(keep)}, delete: {len(to_delete)}')\n",
+    "for i in range(0, len(to_delete), 200):\n",
+    "    batch = to_delete[i:i+200]\n",
+    "    api.delete_files(repo_id=REPO, repo_type='dataset', delete_patterns=batch,\n",
+    "                     commit_message=f'Remove {len(batch)} orphan WAVs (no transcript)')\n",
+    "    print(f'  deleted batch {i//200+1}: {len(batch)} files')\n",
+    "# 2. Clean metadata.csv upload\n",
+    "api.upload_file(path_or_fileobj=os.path.join(LOCAL, 'metadata.csv'),\n",
+    "                path_in_repo='metadata.csv', repo_id=REPO, repo_type='dataset',\n",
+    "                commit_message=f'Clean metadata: {len(keep)} clips, WAVs == rows 1:1')\n",
+    "print('DONE. Verify: https://huggingface.co/datasets/' + REPO)"
+   ]
+  }
+ ],
+ "metadata": {
+  "kernelspec": {"display_name": "Python 3", "name": "python3"},
+  "language_info": {"name": "python"}
+ },
+ "nbformat": 4,
+ "nbformat_minor": 0
+}