메뉴
Privacy-focused offline transcription using local Whisper models. Use when data privacy is critical, offline operation is needed, or avoiding API costs. Runs entirely on your machine.
기본적으로 소스를 먼저 확인하는 Prompt가 선택됩니다. 직접 명령으로 전환하거나 로컬 사본을 다운로드할 수도 있습니다.
설치 여부를 결정하기 전에 SKILL.md와 SkillsMP에 표시된 보조 파일을 읽어 보세요.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
직접 명령은 검토 Prompt를 거치지 않습니다. 실행하기 전에 소스를 확인하세요.
npx skills add https://github.com/CoachSteff/superskills --skill transcriber-local명령은 한 줄로 유지됩니다. 복사하기 전에 가로로 스크롤해 전체 내용을 확인하세요.
로컬 사본을 원하시나요? SkillsMP에서 현재 제공할 수 있는 파일을 다운로드하세요.
SOC 직업 분류 기준
SKILL.md 표시 중
| name | transcriber-local |
| description | Privacy-focused offline transcription using local Whisper models. Use when data privacy is critical, offline operation is needed, or avoiding API costs. Runs entirely on your machine. |
| version | 1.0.0 |
Note: Review PROFILE.md for user-specific model preferences, output formatting, and privacy settings.
Master Briefing: Global brand voice at
~/.superskills/master-briefing.yamlapplies automatically. Skill profile overrides when conflicts exist.
Privacy-first offline transcription using local Whisper models. Perfect for sensitive content, offline workflows, and cost-conscious projects.
LocalTranscriber.py (in src/):
tiny: Fastest, lowest accuracy (~1GB RAM)base: Fast, decent accuracy (~1GB RAM)small: Balanced (~2GB RAM) [Recommended]medium: High accuracy (~5GB RAM)large: Highest accuracy (~10GB RAM)Basic Local Transcription:
from superskills.transcriber_local.src import LocalTranscriber
transcriber = LocalTranscriber(model_size="small")
result = transcriber.transcribe("recording.mp3")
print(result.transcript)
GPU Acceleration:
transcriber = LocalTranscriber(model_size="medium", device="cuda")
result = transcriber.transcribe("video.mp4")
Batch Processing:
files = ["session1.mp3", "session2.mp3", "session3.mp3"]
results = transcriber.transcribe_batch(files)
✅ No Data Transmission: All processing happens locally ✅ GDPR Compliant: No third-party data sharing ✅ Offline Capable: Works without internet ✅ No Usage Limits: Transcribe unlimited content ✅ Zero API Costs: One-time model download only
| Model | Speed (1hr audio) | Accuracy | VRAM |
|---|---|---|---|
| tiny | ~2 min | Good | 1GB |
| base | ~3 min | Better | 1GB |
| small | ~5 min | Great | 2GB |
| medium | ~12 min | Excellent | 5GB |
| large | ~30 min | Best | 10GB |
Times on CPU. GPU acceleration 3-10x faster.
→ author: Converts transcripts into structured content → editor: Polishes transcripts for publication → narrator: Creates audio from edited transcripts (round-trip workflow) → coursepackager: Packages transcripts into training materials
Install Dependencies:
pip install openai-whisper
# Optional GPU support:
pip install openai-whisper[cuda] # NVIDIA
pip install openai-whisper[metal] # Apple Silicon
First Run:
transcriber = LocalTranscriber(model_size="small")
# Model downloads automatically (~500MB for small model)
Use transcriber-local when:
Use transcriber (cloud) when:
Before finalizing transcripts:
Model Download Fails:
Out of Memory:
Slow Performance:
Secure Client Recordings:
# Sensitive coaching sessions - never leave your machine
transcriber = LocalTranscriber(model_size="medium")
result = transcriber.transcribe("client_session.mp3")
Offline Field Work:
# No internet? No problem
transcriber = LocalTranscriber(model_size="base")
results = transcriber.transcribe_batch(["interview1.wav", "interview2.wav"])
Cost-Effective Bulk Processing:
# Process 100 hours of content with zero API costs
transcriber = LocalTranscriber(model_size="small", device="cuda")
for file in large_audio_collection:
result = transcriber.transcribe(file)
After transcription:
authorcoursepackager