[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M6Q4PGYJ35AMFAA8418FVMFS":3,"skill-same:01M6Q4PGYJ35AMFAA8418FVMFS":106},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":15,"categories":16,"keywords":17,"publishedAt":38,"linkOnly":39,"markdownContent":40,"markdownBody":41,"frontMatter":42,"refName":60,"commitSha":61,"skillDirectory":62,"forksCount":63,"repositoryHtmlUrl":64,"repositoryHomepageUrl":6,"ownerHtmlUrl":65,"ownerBlogUrl":6,"ownerDescription":66,"updatedAt":67,"install":68,"resources":81},"01M6Q4PGYJ35AMFAA8418FVMFS","faster-whisper",null,"Local speech-to-text using faster-whisper. 4-6x faster than OpenAI Whisper with identical accuracy; GPU acceleration enables ~20x realtime transcription. Supports standard and distilled models with word-level timestamps.","Kernel8901\u002Fai-agent-skills-classification","Kernel8901","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F72647553?v=4","01M6KV388J2H5PNC8D8P940BJD","01M6KV38CS5B3AY63RKZB410YM",7,"MIT","https:\u002F\u002Fgithub.com\u002FKernel8901\u002Fai-agent-skills-classification\u002Fblob\u002Fmain\u002F语音与转录\u002Ffaster-whisper\u002FSKILL.md",[],[18,19,20,21,22,23,24,25,26,27,28,29,30,31,32,33,34,35,36,37],"faster whisper","whisper","text","distilled","standard","accuracy","timestamps","GPU Acceleration","local","python3","macos","Word","Audio","speech","openai","ml","realtime","windows","linux","openai whisper","2026-11-10 10:12:30",false,"---\nname: faster-whisper\ndescription: Local speech-to-text using faster-whisper. 4-6x faster than OpenAI Whisper with identical accuracy; GPU acceleration enables ~20x realtime transcription. Supports standard and distilled models with word-level timestamps.\nversion: 1.0.4\nauthor: ThePlasmak\nhomepage: https:\u002F\u002Fgithub.com\u002FThePlasmak\u002Ffaster-whisper\ntags: [\"audio\", \"transcription\", \"whisper\", \"speech-to-text\", \"ml\", \"cuda\", \"gpu\"]\nplatforms: [\"windows\", \"linux\", \"macos\", \"wsl2\"]\nmetadata: {\"moltbot\":{\"emoji\":\"?️\",\"requires\":{\"bins\":[\"ffmpeg\",\"python3\"]}}}\n---\n\n# Faster Whisper\n\nLocal speech-to-text using faster-whisper — a CTranslate2 reimplementation of OpenAI's Whisper that runs **4-6x faster** with identical accuracy. With GPU acceleration, expect **~20x realtime** transcription (a 10-minute audio file in ~30 seconds).\n\n## When to Use\n\nUse this skill when you need to:\n- **Transcribe audio\u002Fvideo files** — meetings, interviews, podcasts, lectures, YouTube videos\n- **Convert speech to text locally** — no API costs, works offline (after model download)\n- **Batch process multiple audio files** — efficient for large collections\n- **Generate subtitles\u002Fcaptions** — word-level timestamps available\n- **Multilingual transcription** — supports 99+ languages with auto-detection\n\n**Trigger phrases:** \"transcribe this audio\", \"convert speech to text\", \"what did they say\", \"make a transcript\", \"audio to text\", \"subtitle this video\"\n\n**When NOT to use:**\n- Real-time\u002Fstreaming transcription (use streaming-optimized tools instead)\n- Cloud-only environments without local compute\n- Files \u003C10 seconds where API call latency doesn't matter\n\n## Quick Reference\n\n| Task | Command | Notes |\n|------|---------|-------|\n| **Basic transcription** | `.\u002Fscripts\u002Ftranscribe audio.mp3` | Uses default distil-large-v3 |\n| **Faster English** | `.\u002Fscripts\u002Ftranscribe audio.mp3 --model distil-medium.en --language en` | English-only, 6.8x faster |\n| **Maximum accuracy** | `.\u002Fscripts\u002Ftranscribe audio.mp3 --model large-v3-turbo --beam-size 10` | Slower but best quality |\n| **Word timestamps** | `.\u002Fscripts\u002Ftranscribe audio.mp3 --word-timestamps` | For subtitles\u002Fcaptions |\n| **JSON output** | `.\u002Fscripts\u002Ftranscribe audio.mp3 --json -o output.json` | Programmatic access |\n| **Multilingual** | `.\u002Fscripts\u002Ftranscribe audio.mp3 --model large-v3-turbo` | Auto-detects language |\n| **Remove silence** | `.\u002Fscripts\u002Ftranscribe audio.mp3 --vad` | Voice activity detection |\n\n## Model Selection\n\nChoose the right model for your needs:\n\n```dot\ndigraph model_selection {\n    rankdir=LR;\n    node [shape=box, style=rounded];\n\n    start [label=\"Start\", shape=doublecircle];\n    need_accuracy [label=\"Need maximum\\naccuracy?\", shape=diamond];\n    multilingual [label=\"Multilingual\\ncontent?\", shape=diamond];\n    resource_constrained [label=\"Resource\\nconstraints?\", shape=diamond];\n\n    large_v3 [label=\"large-v3\\nor\\nlarge-v3-turbo\", style=\"rounded,filled\", fillcolor=lightblue];\n    large_turbo [label=\"large-v3-turbo\", style=\"rounded,filled\", fillcolor=lightblue];\n    distil_large [label=\"distil-large-v3\\n(default)\", style=\"rounded,filled\", fillcolor=lightgreen];\n    distil_medium [label=\"distil-medium.en\", style=\"rounded,filled\", fillcolor=lightyellow];\n    distil_small [label=\"distil-small.en\", style=\"rounded,filled\", fillcolor=lightyellow];\n\n    start -> need_accuracy;\n    need_accuracy -> large_v3 [label=\"yes\"];\n    need_accuracy -> multilingual [label=\"no\"];\n    multilingual -> large_turbo [label=\"yes\"];\n    multilingual -> resource_constrained [label=\"no (English)\"];\n    resource_constrained -> distil_small [label=\"mobile\u002Fedge\"];\n    resource_constrained -> distil_medium [label=\"some limits\"];\n    resource_constrained -> distil_large [label=\"no\"];\n}\n```\n\n### Model Table\n\n#### Standard Models (Full Whisper)\n| Model | Size | Speed | Accuracy | Use Case |\n|-------|------|-------|----------|----------|\n| `tiny` \u002F `tiny.en` | 39M | Fastest | Basic | Quick drafts |\n| `base` \u002F `base.en` | 74M | Very fast | Good | General use |\n| `small` \u002F `small.en` | 244M | Fast | Better | Most tasks |\n| `medium` \u002F `medium.en` | 769M | Moderate | High | Quality transcription |\n| `large-v1\u002Fv2\u002Fv3` | 1.5GB | Slower | Best | Maximum accuracy |\n| **`large-v3-turbo`** | 809M | Fast | Excellent | **Recommended for accuracy** |\n\n#### Distilled Models (~6x Faster, ~1% WER difference)\n| Model | Size | Speed vs Standard | Accuracy | Use Case |\n|-------|------|-------------------|----------|----------|\n| **`distil-large-v3`** | 756M | ~6.3x faster | 9.7% WER | **Default, best balance** |\n| `distil-large-v2` | 756M | ~5.8x faster | 10.1% WER | Fallback |\n| `distil-medium.en` | 394M | ~6.8x faster | 11.1% WER | English-only, resource-constrained |\n| `distil-small.en` | 166M | ~5.6x faster | 12.1% WER | Mobile\u002Fedge devices |\n\n`.en` models are English-only and slightly faster\u002Fbetter for English content.\n\n## Setup\n\n### Linux \u002F macOS \u002F WSL2\n```bash\n# Run the setup script (creates venv, installs deps, auto-detects GPU)\n.\u002Fsetup.sh\n```\n\n### Windows (Native)\n```powershell\n# Run from PowerShell (auto-installs Python & ffmpeg if missing via winget)\n.\\setup.ps1\n```\n\nThe Windows setup script will:\n- Auto-install Python 3.12 via winget if not found\n- Auto-install ffmpeg via winget if not found\n- Detect NVIDIA GPU and install CUDA-enabled PyTorch\n- Create venv and install all dependencies\n\nRequirements:\n- **Linux\u002FmacOS\u002FWSL2**: Python 3.10+, ffmpeg\n- **Windows**: Nothing! Setup auto-installs prerequisites via winget\n\n### Platform Support\n\n| Platform | Acceleration | Speed | Auto-Install |\n|----------|-------------|-------|--------------|\n| **Windows + NVIDIA GPU** | CUDA | ~20x realtime ? | ✅ Full |\n| **Linux + NVIDIA GPU** | CUDA | ~20x realtime ? | Manual prereqs |\n| **WSL2 + NVIDIA GPU** | CUDA | ~20x realtime ? | Manual prereqs |\n| macOS Apple Silicon | CPU* | ~3-5x realtime | Manual prereqs |\n| macOS Intel | CPU | ~1-2x realtime | Manual prereqs |\n| Windows (no GPU) | CPU | ~1x realtime | ✅ Full |\n| Linux (no GPU) | CPU | ~1x realtime | Manual prereqs |\n\n*faster-whisper uses CTranslate2 which is CPU-only on macOS, but Apple Silicon is fast enough for practical use.\n\n### GPU Support (IMPORTANT!)\n\nThe setup script auto-detects your GPU and installs PyTorch with CUDA. **Always use GPU if available** — CPU transcription is extremely slow.\n\n| Hardware | Speed | 9-min video |\n|----------|-------|-------------|\n| RTX 3070 (GPU) | ~20x realtime | ~27 sec |\n| CPU (int8) | ~0.3x realtime | ~30 min |\n\nIf setup didn't detect your GPU, manually install PyTorch with CUDA:\n\n**Linux\u002FmacOS\u002FWSL2:**\n```bash\n# For CUDA 12.x\nuv pip install --python .venv\u002Fbin\u002Fpython torch --index-url https:\u002F\u002Fdownload.pytorch.org\u002Fwhl\u002Fcu121\n\n# For CUDA 11.x\nuv pip install --python .venv\u002Fbin\u002Fpython torch --index-url https:\u002F\u002Fdownload.pytorch.org\u002Fwhl\u002Fcu118\n```\n\n**Windows:**\n```powershell\n# For CUDA 12.x\n.venv\\Scripts\\pip install torch --index-url https:\u002F\u002Fdownload.pytorch.org\u002Fwhl\u002Fcu121\n\n# For CUDA 11.x\n.venv\\Scripts\\pip install torch --index-url https:\u002F\u002Fdownload.pytorch.org\u002Fwhl\u002Fcu118\n```\n\n- **Windows users**: Ensure you have [NVIDIA drivers](https:\u002F\u002Fwww.nvidia.com\u002Fdownload\u002Findex.aspx) installed\n- **WSL2 users**: Ensure you have the [NVIDIA CUDA drivers for WSL](https:\u002F\u002Fdocs.nvidia.com\u002Fcuda\u002Fwsl-user-guide\u002F) installed on Windows\n\n## Usage\n\n**Linux\u002FmacOS\u002FWSL2:**\n```bash\n# Basic transcription\n.\u002Fscripts\u002Ftranscribe audio.mp3\n\n# With specific model\n.\u002Fscripts\u002Ftranscribe audio.wav --model large-v3-turbo\n\n# With word timestamps\n.\u002Fscripts\u002Ftranscribe audio.mp3 --word-timestamps\n\n# Specify language (faster than auto-detect)\n.\u002Fscripts\u002Ftranscribe audio.mp3 --language en\n\n# JSON output\n.\u002Fscripts\u002Ftranscribe audio.mp3 --json\n```\n\n**Windows (cmd or PowerShell):**\n```powershell\n# Basic transcription\n.\\scripts\\transcribe.cmd audio.mp3\n\n# With specific model\n.\\scripts\\transcribe.cmd audio.wav --model large-v3-turbo\n\n# With word timestamps (PowerShell native syntax also works)\n.\\scripts\\transcribe.ps1 audio.mp3 -WordTimestamps\n\n# JSON output\n.\\scripts\\transcribe.cmd audio.mp3 --json\n```\n\n## Options\n\n```\n--model, -m        Model name (default: distil-large-v3)\n--language, -l     Language code (e.g., en, es, fr - auto-detect if omitted)\n--word-timestamps  Include word-level timestamps\n--beam-size        Beam search size (default: 5, higher = more accurate but slower)\n--vad              Enable voice activity detection (removes silence)\n--json, -j         Output as JSON\n--output, -o       Save transcript to file\n--device           cpu or cuda (auto-detected)\n--compute-type     int8, float16, float32 (default: auto-optimized)\n--quiet, -q        Suppress progress messages\n```\n\n## Examples\n\n```bash\n# Transcribe YouTube audio (after extraction with yt-dlp)\nyt-dlp -x --audio-format mp3 \u003CURL> -o audio.mp3\n.\u002Fscripts\u002Ftranscribe audio.mp3\n\n# Batch transcription with JSON output\nfor file in *.mp3; do\n  .\u002Fscripts\u002Ftranscribe \"$file\" --json > \"${file%.mp3}.json\"\ndone\n\n# High-accuracy transcription with larger beam size\n.\u002Fscripts\u002Ftranscribe audio.mp3 \\\n  --model large-v3-turbo --beam-size 10 --word-timestamps\n\n# Fast English-only transcription\n.\u002Fscripts\u002Ftranscribe audio.mp3 \\\n  --model distil-medium.en --language en\n\n# Transcribe with VAD (removes silence)\n.\u002Fscripts\u002Ftranscribe audio.mp3 --vad\n```\n\n## Common Mistakes\n\n| Mistake | Problem | Solution |\n|---------|---------|----------|\n| **Using CPU when GPU available** | 10-20x slower transcription | Check `nvidia-smi` on Windows\u002FLinux; verify CUDA installation |\n| **Not specifying language** | Wastes time auto-detecting on known content | Use `--language en` when you know the language |\n| **Using wrong model** | Unnecessary slowness or poor accuracy | Default `distil-large-v3` is excellent; only use `large-v3` if accuracy issues |\n| **Ignoring distilled models** | Missing 6x speedup with \u003C1% accuracy loss | Try `distil-large-v3` before reaching for standard models |\n| **Forgetting ffmpeg** | Setup fails or audio can't be processed | Setup script handles this; manual installs need ffmpeg separately |\n| **Out of memory errors** | Model too large for available VRAM\u002FRAM | Use smaller model or `--compute-type int8` |\n| **Over-engineering beam size** | Diminishing returns past beam-size 5-7 | Default 5 is fine; try 10 for critical transcripts |\n\n## Performance Notes\n\n- **First run**: Downloads model to `~\u002F.cache\u002Fhuggingface\u002F` (one-time)\n- **GPU**: Automatically uses CUDA if available (~10-20x faster)\n- **Quantization**: INT8 used on CPU for ~4x speedup with minimal accuracy loss\n- **Memory**:\n  - `distil-large-v3`: ~2GB RAM \u002F ~1GB VRAM\n  - `large-v3-turbo`: ~4GB RAM \u002F ~2GB VRAM\n  - `tiny\u002Fbase`: \u003C1GB RAM\n\n## Why faster-whisper?\n\n- **Speed**: ~4-6x faster than OpenAI's original Whisper\n- **Accuracy**: Identical (uses same model weights)\n- **Efficiency**: Lower memory usage via quantization\n- **Production-ready**: Stable C++ backend (CTranslate2)\n- **Distilled models**: ~6x faster with \u003C1% accuracy loss\n\n## Troubleshooting\n\n**\"CUDA not available — using CPU\"**: Install PyTorch with CUDA (see GPU Support above)\n**Setup fails**: Make sure Python 3.10+ is installed\n**Out of memory**: Use smaller model or `--compute-type int8`\n**Slow on CPU**: Expected — use GPU for practical transcription\n**Model download fails**: Check `~\u002F.cache\u002Fhuggingface\u002F` permissions (Linux\u002FmacOS) or `%USERPROFILE%\\.cache\\huggingface\\` (Windows)\n\n### Windows-Specific\n\n**\"winget not found\"**: Install [App Installer](https:\u002F\u002Fapps.microsoft.com\u002Fdetail\u002F9NBLGGH4NNS1) from Microsoft Store, or install Python\u002Fffmpeg manually\n**\"Python not in PATH after install\"**: Close and reopen your terminal, then run `setup.ps1` again\n**PowerShell execution policy error**: Run `Set-ExecutionPolicy -Scope CurrentUser -ExecutionPolicy RemoteSigned` or use `transcribe.cmd` instead\n**nvidia-smi not found but have GPU**: Install [NVIDIA drivers](https:\u002F\u002Fwww.nvidia.com\u002Fdownload\u002Findex.aspx) — the Game Ready or Studio drivers include nvidia-smi\n\n## References\n\n- [faster-whisper GitHub](https:\u002F\u002Fgithub.com\u002FSYSTRAN\u002Ffaster-whisper)\n- [Distil-Whisper Paper](https:\u002F\u002Farxiv.org\u002Fabs\u002F2311.00430)\n- [HuggingFace Models](https:\u002F\u002Fhuggingface.co\u002Fcollections\u002FSystran\u002Ffaster-whisper)\n","# Faster Whisper\n\nLocal speech-to-text using faster-whisper — a CTranslate2 reimplementation of OpenAI's Whisper that runs **4-6x faster** with identical accuracy. With GPU acceleration, expect **~20x realtime** transcription (a 10-minute audio file in ~30 seconds).\n\n## When to Use\n\nUse this skill when you need to:\n- **Transcribe audio\u002Fvideo files** — meetings, interviews, podcasts, lectures, YouTube videos\n- **Convert speech to text locally** — no API costs, works offline (after model download)\n- **Batch process multiple audio files** — efficient for large collections\n- **Generate subtitles\u002Fcaptions** — word-level timestamps available\n- **Multilingual transcription** — supports 99+ languages with auto-detection\n\n**Trigger phrases:** \"transcribe this audio\", \"convert speech to text\", \"what did they say\", \"make a transcript\", \"audio to text\", \"subtitle this video\"\n\n**When NOT to use:**\n- Real-time\u002Fstreaming transcription (use streaming-optimized tools instead)\n- Cloud-only environments without local compute\n- Files \u003C10 seconds where API call latency doesn't matter\n\n## Quick Reference\n\n| Task | Command | Notes |\n|------|---------|-------|\n| **Basic transcription** | `.\u002Fscripts\u002Ftranscribe audio.mp3` | Uses default distil-large-v3 |\n| **Faster English** | `.\u002Fscripts\u002Ftranscribe audio.mp3 --model distil-medium.en --language en` | English-only, 6.8x faster |\n| **Maximum accuracy** | `.\u002Fscripts\u002Ftranscribe audio.mp3 --model large-v3-turbo --beam-size 10` | Slower but best quality |\n| **Word timestamps** | `.\u002Fscripts\u002Ftranscribe audio.mp3 --word-timestamps` | For subtitles\u002Fcaptions |\n| **JSON output** | `.\u002Fscripts\u002Ftranscribe audio.mp3 --json -o output.json` | Programmatic access |\n| **Multilingual** | `.\u002Fscripts\u002Ftranscribe audio.mp3 --model large-v3-turbo` | Auto-detects language |\n| **Remove silence** | `.\u002Fscripts\u002Ftranscribe audio.mp3 --vad` | Voice activity detection |\n\n## Model Selection\n\nChoose the right model for your needs:\n\n```dot\ndigraph model_selection {\n    rankdir=LR;\n    node [shape=box, style=rounded];\n\n    start [label=\"Start\", shape=doublecircle];\n    need_accuracy [label=\"Need maximum\\naccuracy?\", shape=diamond];\n    multilingual [label=\"Multilingual\\ncontent?\", shape=diamond];\n    resource_constrained [label=\"Resource\\nconstraints?\", shape=diamond];\n\n    large_v3 [label=\"large-v3\\nor\\nlarge-v3-turbo\", style=\"rounded,filled\", fillcolor=lightblue];\n    large_turbo [label=\"large-v3-turbo\", style=\"rounded,filled\", fillcolor=lightblue];\n    distil_large [label=\"distil-large-v3\\n(default)\", style=\"rounded,filled\", fillcolor=lightgreen];\n    distil_medium [label=\"distil-medium.en\", style=\"rounded,filled\", fillcolor=lightyellow];\n    distil_small [label=\"distil-small.en\", style=\"rounded,filled\", fillcolor=lightyellow];\n\n    start -> need_accuracy;\n    need_accuracy -> large_v3 [label=\"yes\"];\n    need_accuracy -> multilingual [label=\"no\"];\n    multilingual -> large_turbo [label=\"yes\"];\n    multilingual -> resource_constrained [label=\"no (English)\"];\n    resource_constrained -> distil_small [label=\"mobile\u002Fedge\"];\n    resource_constrained -> distil_medium [label=\"some limits\"];\n    resource_constrained -> distil_large [label=\"no\"];\n}\n```\n\n### Model Table\n\n#### Standard Models (Full Whisper)\n| Model | Size | Speed | Accuracy | Use Case |\n|-------|------|-------|----------|----------|\n| `tiny` \u002F `tiny.en` | 39M | Fastest | Basic | Quick drafts |\n| `base` \u002F `base.en` | 74M | Very fast | Good | General use |\n| `small` \u002F `small.en` | 244M | Fast | Better | Most tasks |\n| `medium` \u002F `medium.en` | 769M | Moderate | High | Quality transcription |\n| `large-v1\u002Fv2\u002Fv3` | 1.5GB | Slower | Best | Maximum accuracy |\n| **`large-v3-turbo`** | 809M | Fast | Excellent | **Recommended for accuracy** |\n\n#### Distilled Models (~6x Faster, ~1% WER difference)\n| Model | Size | Speed vs Standard | Accuracy | Use Case |\n|-------|------|-------------------|----------|----------|\n| **`distil-large-v3`** | 756M | ~6.3x faster | 9.7% WER | **Default, best balance** |\n| `distil-large-v2` | 756M | ~5.8x faster | 10.1% WER | Fallback |\n| `distil-medium.en` | 394M | ~6.8x faster | 11.1% WER | English-only, resource-constrained |\n| `distil-small.en` | 166M | ~5.6x faster | 12.1% WER | Mobile\u002Fedge devices |\n\n`.en` models are English-only and slightly faster\u002Fbetter for English content.\n\n## Setup\n\n### Linux \u002F macOS \u002F WSL2\n```bash\n# Run the setup script (creates venv, installs deps, auto-detects GPU)\n.\u002Fsetup.sh\n```\n\n### Windows (Native)\n```powershell\n# Run from PowerShell (auto-installs Python & ffmpeg if missing via winget)\n.\\setup.ps1\n```\n\nThe Windows setup script will:\n- Auto-install Python 3.12 via winget if not found\n- Auto-install ffmpeg via winget if not found\n- Detect NVIDIA GPU and install CUDA-enabled PyTorch\n- Create venv and install all dependencies\n\nRequirements:\n- **Linux\u002FmacOS\u002FWSL2**: Python 3.10+, ffmpeg\n- **Windows**: Nothing! Setup auto-installs prerequisites via winget\n\n### Platform Support\n\n| Platform | Acceleration | Speed | Auto-Install |\n|----------|-------------|-------|--------------|\n| **Windows + NVIDIA GPU** | CUDA | ~20x realtime ? | ✅ Full |\n| **Linux + NVIDIA GPU** | CUDA | ~20x realtime ? | Manual prereqs |\n| **WSL2 + NVIDIA GPU** | CUDA | ~20x realtime ? | Manual prereqs |\n| macOS Apple Silicon | CPU* | ~3-5x realtime | Manual prereqs |\n| macOS Intel | CPU | ~1-2x realtime | Manual prereqs |\n| Windows (no GPU) | CPU | ~1x realtime | ✅ Full |\n| Linux (no GPU) | CPU | ~1x realtime | Manual prereqs |\n\n*faster-whisper uses CTranslate2 which is CPU-only on macOS, but Apple Silicon is fast enough for practical use.\n\n### GPU Support (IMPORTANT!)\n\nThe setup script auto-detects your GPU and installs PyTorch with CUDA. **Always use GPU if available** — CPU transcription is extremely slow.\n\n| Hardware | Speed | 9-min video |\n|----------|-------|-------------|\n| RTX 3070 (GPU) | ~20x realtime | ~27 sec |\n| CPU (int8) | ~0.3x realtime | ~30 min |\n\nIf setup didn't detect your GPU, manually install PyTorch with CUDA:\n\n**Linux\u002FmacOS\u002FWSL2:**\n```bash\n# For CUDA 12.x\nuv pip install --python .venv\u002Fbin\u002Fpython torch --index-url https:\u002F\u002Fdownload.pytorch.org\u002Fwhl\u002Fcu121\n\n# For CUDA 11.x\nuv pip install --python .venv\u002Fbin\u002Fpython torch --index-url https:\u002F\u002Fdownload.pytorch.org\u002Fwhl\u002Fcu118\n```\n\n**Windows:**\n```powershell\n# For CUDA 12.x\n.venv\\Scripts\\pip install torch --index-url https:\u002F\u002Fdownload.pytorch.org\u002Fwhl\u002Fcu121\n\n# For CUDA 11.x\n.venv\\Scripts\\pip install torch --index-url https:\u002F\u002Fdownload.pytorch.org\u002Fwhl\u002Fcu118\n```\n\n- **Windows users**: Ensure you have [NVIDIA drivers](https:\u002F\u002Fwww.nvidia.com\u002Fdownload\u002Findex.aspx) installed\n- **WSL2 users**: Ensure you have the [NVIDIA CUDA drivers for WSL](https:\u002F\u002Fdocs.nvidia.com\u002Fcuda\u002Fwsl-user-guide\u002F) installed on Windows\n\n## Usage\n\n**Linux\u002FmacOS\u002FWSL2:**\n```bash\n# Basic transcription\n.\u002Fscripts\u002Ftranscribe audio.mp3\n\n# With specific model\n.\u002Fscripts\u002Ftranscribe audio.wav --model large-v3-turbo\n\n# With word timestamps\n.\u002Fscripts\u002Ftranscribe audio.mp3 --word-timestamps\n\n# Specify language (faster than auto-detect)\n.\u002Fscripts\u002Ftranscribe audio.mp3 --language en\n\n# JSON output\n.\u002Fscripts\u002Ftranscribe audio.mp3 --json\n```\n\n**Windows (cmd or PowerShell):**\n```powershell\n# Basic transcription\n.\\scripts\\transcribe.cmd audio.mp3\n\n# With specific model\n.\\scripts\\transcribe.cmd audio.wav --model large-v3-turbo\n\n# With word timestamps (PowerShell native syntax also works)\n.\\scripts\\transcribe.ps1 audio.mp3 -WordTimestamps\n\n# JSON output\n.\\scripts\\transcribe.cmd audio.mp3 --json\n```\n\n## Options\n\n```\n--model, -m        Model name (default: distil-large-v3)\n--language, -l     Language code (e.g., en, es, fr - auto-detect if omitted)\n--word-timestamps  Include word-level timestamps\n--beam-size        Beam search size (default: 5, higher = more accurate but slower)\n--vad              Enable voice activity detection (removes silence)\n--json, -j         Output as JSON\n--output, -o       Save transcript to file\n--device           cpu or cuda (auto-detected)\n--compute-type     int8, float16, float32 (default: auto-optimized)\n--quiet, -q        Suppress progress messages\n```\n\n## Examples\n\n```bash\n# Transcribe YouTube audio (after extraction with yt-dlp)\nyt-dlp -x --audio-format mp3 \u003CURL> -o audio.mp3\n.\u002Fscripts\u002Ftranscribe audio.mp3\n\n# Batch transcription with JSON output\nfor file in *.mp3; do\n  .\u002Fscripts\u002Ftranscribe \"$file\" --json > \"${file%.mp3}.json\"\ndone\n\n# High-accuracy transcription with larger beam size\n.\u002Fscripts\u002Ftranscribe audio.mp3 \\\n  --model large-v3-turbo --beam-size 10 --word-timestamps\n\n# Fast English-only transcription\n.\u002Fscripts\u002Ftranscribe audio.mp3 \\\n  --model distil-medium.en --language en\n\n# Transcribe with VAD (removes silence)\n.\u002Fscripts\u002Ftranscribe audio.mp3 --vad\n```\n\n## Common Mistakes\n\n| Mistake | Problem | Solution |\n|---------|---------|----------|\n| **Using CPU when GPU available** | 10-20x slower transcription | Check `nvidia-smi` on Windows\u002FLinux; verify CUDA installation |\n| **Not specifying language** | Wastes time auto-detecting on known content | Use `--language en` when you know the language |\n| **Using wrong model** | Unnecessary slowness or poor accuracy | Default `distil-large-v3` is excellent; only use `large-v3` if accuracy issues |\n| **Ignoring distilled models** | Missing 6x speedup with \u003C1% accuracy loss | Try `distil-large-v3` before reaching for standard models |\n| **Forgetting ffmpeg** | Setup fails or audio can't be processed | Setup script handles this; manual installs need ffmpeg separately |\n| **Out of memory errors** | Model too large for available VRAM\u002FRAM | Use smaller model or `--compute-type int8` |\n| **Over-engineering beam size** | Diminishing returns past beam-size 5-7 | Default 5 is fine; try 10 for critical transcripts |\n\n## Performance Notes\n\n- **First run**: Downloads model to `~\u002F.cache\u002Fhuggingface\u002F` (one-time)\n- **GPU**: Automatically uses CUDA if available (~10-20x faster)\n- **Quantization**: INT8 used on CPU for ~4x speedup with minimal accuracy loss\n- **Memory**:\n  - `distil-large-v3`: ~2GB RAM \u002F ~1GB VRAM\n  - `large-v3-turbo`: ~4GB RAM \u002F ~2GB VRAM\n  - `tiny\u002Fbase`: \u003C1GB RAM\n\n## Why faster-whisper?\n\n- **Speed**: ~4-6x faster than OpenAI's original Whisper\n- **Accuracy**: Identical (uses same model weights)\n- **Efficiency**: Lower memory usage via quantization\n- **Production-ready**: Stable C++ backend (CTranslate2)\n- **Distilled models**: ~6x faster with \u003C1% accuracy loss\n\n## Troubleshooting\n\n**\"CUDA not available — using CPU\"**: Install PyTorch with CUDA (see GPU Support above)\n**Setup fails**: Make sure Python 3.10+ is installed\n**Out of memory**: Use smaller model or `--compute-type int8`\n**Slow on CPU**: Expected — use GPU for practical transcription\n**Model download fails**: Check `~\u002F.cache\u002Fhuggingface\u002F` permissions (Linux\u002FmacOS) or `%USERPROFILE%\\.cache\\huggingface\\` (Windows)\n\n### Windows-Specific\n\n**\"winget not found\"**: Install [App Installer](https:\u002F\u002Fapps.microsoft.com\u002Fdetail\u002F9NBLGGH4NNS1) from Microsoft Store, or install Python\u002Fffmpeg manually\n**\"Python not in PATH after install\"**: Close and reopen your terminal, then run `setup.ps1` again\n**PowerShell execution policy error**: Run `Set-ExecutionPolicy -Scope CurrentUser -ExecutionPolicy RemoteSigned` or use `transcribe.cmd` instead\n**nvidia-smi not found but have GPU**: Install [NVIDIA drivers](https:\u002F\u002Fwww.nvidia.com\u002Fdownload\u002Findex.aspx) — the Game Ready or Studio drivers include nvidia-smi\n\n## References\n\n- [faster-whisper GitHub](https:\u002F\u002Fgithub.com\u002FSYSTRAN\u002Ffaster-whisper)\n- [Distil-Whisper Paper](https:\u002F\u002Farxiv.org\u002Fabs\u002F2311.00430)\n- [HuggingFace Models](https:\u002F\u002Fhuggingface.co\u002Fcollections\u002FSystran\u002Ffaster-whisper)\n",{"name":5,"tags":43,"author":49,"version":50,"homepage":51,"metadata":52,"platforms":58,"description":7},[44,45,19,46,33,47,48],"audio","transcription","speech-to-text","cuda","gpu","ThePlasmak","1.0.4","https:\u002F\u002Fgithub.com\u002FThePlasmak\u002Ffaster-whisper",{"moltbot":53},{"emoji":54,"requires":55},"?️",{"bins":56},[57,27],"ffmpeg",[35,36,28,59],"wsl2","main","f4cb2d49f7667e79ef7804f50396d5aaffdf1f56","语音与转录\u002Ffaster-whisper",1,"https:\u002F\u002Fgithub.com\u002FKernel8901\u002Fai-agent-skills-classification","https:\u002F\u002Fgithub.com\u002FKernel8901","a little little little dobe in life but science geek","2026-04-05 16:43:56",{"source":8,"skillName":5,"commands":69,"prompt":79,"zipAvailable":80},[70,73,76],{"tool":71,"command":72},"npx","npx skills add Kernel8901\u002Fai-agent-skills-classification --skill faster-whisper",{"tool":74,"command":75},"bunx","bunx skills add Kernel8901\u002Fai-agent-skills-classification --skill faster-whisper",{"tool":77,"command":78},"pnpm","pnpm dlx skills add Kernel8901\u002Fai-agent-skills-classification --skill faster-whisper","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FKernel8901\u002Fai-agent-skills-classification\u002Fblob\u002Fmain\u002F语音与转录\u002Ffaster-whisper\u002FSKILL.md",true,[82,88,92,98,102],{"path":83,"resourceType":84,"resourceTypeDesc":85,"sizeBytes":86,"sourceHtmlUrl":87},"_meta.json","REFERENCE","参考文档",638,"https:\u002F\u002Fgithub.com\u002FKernel8901\u002Fai-agent-skills-classification\u002Fblob\u002Fmain\u002F语音与转录\u002Ffaster-whisper\u002F_meta.json",{"path":89,"resourceType":84,"resourceTypeDesc":85,"sizeBytes":90,"sourceHtmlUrl":91},"requirements.txt",117,"https:\u002F\u002Fgithub.com\u002FKernel8901\u002Fai-agent-skills-classification\u002Fblob\u002Fmain\u002F语音与转录\u002Ffaster-whisper\u002Frequirements.txt",{"path":93,"resourceType":94,"resourceTypeDesc":95,"sizeBytes":96,"sourceHtmlUrl":97},"scripts\u002Ftranscribe.py","SCRIPT","脚本",6292,"https:\u002F\u002Fgithub.com\u002FKernel8901\u002Fai-agent-skills-classification\u002Fblob\u002Fmain\u002F语音与转录\u002Ffaster-whisper\u002Fscripts\u002Ftranscribe.py",{"path":99,"resourceType":94,"resourceTypeDesc":95,"sizeBytes":100,"sourceHtmlUrl":101},"setup.sh",5223,"https:\u002F\u002Fgithub.com\u002FKernel8901\u002Fai-agent-skills-classification\u002Fblob\u002Fmain\u002F语音与转录\u002Ffaster-whisper\u002Fsetup.sh",{"path":103,"resourceType":84,"resourceTypeDesc":85,"sizeBytes":104,"sourceHtmlUrl":105},"skill.json",357,"https:\u002F\u002Fgithub.com\u002FKernel8901\u002Fai-agent-skills-classification\u002Fblob\u002Fmain\u002F语音与转录\u002Ffaster-whisper\u002Fskill.json",[107,131,167,178,188,217],{"skillNo":108,"name":109,"nameZh":6,"description":110,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":111,"categories":112,"keywords":113,"publishedAt":38},"01M6Q5R6WZ06RRB9RQTW54B34B","zyfai","Earn yield on any Ethereum wallet on Base, Arbitrum, and Plasma. Use when a user wants passive DeFi yield on their funds. Deploys a non-custodial subaccount (Safe) linked to their EOA, enables automated yield optimization, and lets them deposit\u002Fwithdraw anytime.","https:\u002F\u002Fgithub.com\u002FKernel8901\u002Fai-agent-skills-classification\u002Fblob\u002Fmain\u002F金融\u002Fzyfai-sdk\u002FSKILL.md",[],[109,114,115,116,117,118,119,120,121,122,123,124,125,126,127,128,129,130],"safe","ethereum","defi","Optimization","user","use","Wallet","base","deposit","withdraw","yield","arbitrum","non-custodial","passive","defi yield","earn","yield optimization",{"skillNo":132,"name":133,"nameZh":6,"description":6,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":134,"categories":135,"keywords":149,"publishedAt":38},"01M6Q5R337B5JSNER4P4YS03TN","zerion-api","https:\u002F\u002Fgithub.com\u002FKernel8901\u002Fai-agent-skills-classification\u002Fblob\u002Fmain\u002F金融\u002Fzerion-api\u002FSKILL.md",[136,141,144],{"categoryNo":137,"categoryName":138,"taxonomyType":139,"taxonomyTypeDesc":140},"01JBSEEDCAT000000000000010","后端与接口","SCENARIO","使用场景",{"categoryNo":142,"categoryName":143,"taxonomyType":139,"taxonomyTypeDesc":140},"01JBSEEDCAT000000000000023","Skill与智能体开发",{"categoryNo":145,"categoryName":146,"taxonomyType":147,"taxonomyTypeDesc":148},"01JBSEEDCAT000000000000030","开发工程师","PROFESSION","适用职业",[150,151,152,153,154,155,156,116,157,120,158,159,160,161,162,163,164,165,166],"api","zerion api","zerion","MCP","mcp server","server","blockchain","transactions","crypto wallet","crypto","evm","solana","query","across","positions","token","prices",{"skillNo":168,"name":169,"nameZh":6,"description":6,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":170,"categories":171,"keywords":175,"publishedAt":38},"01M6Q5QZFYTDAZMZ27BN5VZXAR","zerion-api-skill","https:\u002F\u002Fgithub.com\u002FKernel8901\u002Fai-agent-skills-classification\u002Fblob\u002Fmain\u002F金融\u002Fzerion-api-skill\u002FSKILL.md",[172,173,174],{"categoryNo":137,"categoryName":138,"taxonomyType":139,"taxonomyTypeDesc":140},{"categoryNo":142,"categoryName":143,"taxonomyType":139,"taxonomyTypeDesc":140},{"categoryNo":145,"categoryName":146,"taxonomyType":147,"taxonomyTypeDesc":148},[150,176,177,151,152,153,154,155,156,116,157,120,159,160,161,162,163,164,165,166],"api skill","zerion api skill",{"skillNo":179,"name":180,"nameZh":6,"description":6,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":181,"categories":182,"keywords":186,"publishedAt":38},"01M6Q5QW17QW98RA3FE45E4RD8","zerion-api-mcp","https:\u002F\u002Fgithub.com\u002FKernel8901\u002Fai-agent-skills-classification\u002Fblob\u002Fmain\u002F金融\u002Fzerion-api-mcp\u002FSKILL.md",[183,184,185],{"categoryNo":137,"categoryName":138,"taxonomyType":139,"taxonomyTypeDesc":140},{"categoryNo":142,"categoryName":143,"taxonomyType":139,"taxonomyTypeDesc":140},{"categoryNo":145,"categoryName":146,"taxonomyType":147,"taxonomyTypeDesc":148},[153,150,187,151,152,154,155,116,157,120,158,159,160,161,162,163,164,165,166],"zerion api mcp",{"skillNo":189,"name":190,"nameZh":6,"description":191,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":192,"categories":193,"keywords":197,"publishedAt":38},"01M6Q5QQYZB7A501BDEQWKGSJT","yahoo-finance-forex","Fetches real-time FOREX news and market data from Yahoo Finance for major currency pairs (EUR\u002FUSD, GBP\u002FUSD, USD\u002FJPY, etc.). Analyzes sentiment and provides trading context.","https:\u002F\u002Fgithub.com\u002FKernel8901\u002Fai-agent-skills-classification\u002Fblob\u002Fmain\u002F金融\u002Fyahoo-finance-forex\u002FSKILL.md",[194],{"categoryNo":195,"categoryName":196,"taxonomyType":147,"taxonomyTypeDesc":148},"01JBSEEDCAT000000000000040","财务",[198,199,200,201,202,203,204,205,206,27,207,208,209,210,211,212,213,214,215,216],"finance","yahoo finance","forex","yahoo","yahoo finance forex","github","https","sentiment","real-time","install","Market","usd","news","time","currency","gbp","pip","market data","for",{"skillNo":218,"name":219,"nameZh":6,"description":220,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":221,"categories":222,"keywords":224,"publishedAt":38},"01M6Q5QM6VS84WZVM7BKXXPF32","yahoo-finance","This skill should be used when the user asks to \"get stock prices\", \"check stock quotes\", \"look up earnings\", \"get financial data\", \"find trending stocks\", or needs stock market data from Yahoo Finance.","https:\u002F\u002Fgithub.com\u002FKernel8901\u002Fai-agent-skills-classification\u002Fblob\u002Fmain\u002F金融\u002Fyahoo-finance-cli\u002FSKILL.md",[223],{"categoryNo":195,"categoryName":196,"taxonomyType":147,"taxonomyTypeDesc":148},[198,199,201,225,226,227,228,229,230,208,26,231,232,233,234,235,236,237,166,215],"cli","homebrew","find","check","brew","up","get","stock","link","trending","sf","quotes","ln"]