Lecture2notes — Productivity skill for Claude Code
Lecture recordings to structured notes, synced viewer and course hub.
How to install Lecture2notes
This entry records only its repository, not the path inside it, so there is no
exact command to give. Open jieyu166/lecture2notes and copy the folder into
~/.claude/skills/, or the file into ~/.claude/agents/.
What Lecture2notes does
Lecture recordings to structured notes, synced viewer and course hub. Local ASR (Breeze-ASR-25 / faster-whisper / whisper.cpp / Qwen3-ASR), slide capture, OCR, note-writing guideline. Claude Code / Codex skill + plain CLI.
Alternatives in Productivity
- Claudekit Problem Solving — Structured problem-solving tools 1.9k ★
- Agentlas Hub — Staff a task only from public Agentlas Hub agents 1.1k ★
- Maestro Knowhow — Intent-driven knowhow precipitation — describe what you want to capture (记一个关于X的决策 / 保存这段代码模板 / 写个部署配方 / 存个调试技 530 ★
README
lecture2notes
把講座錄影變成可讀的筆記、時間同步的 viewer 與課程首頁。本機 ASR 轉錄(Breeze-ASR-25 / faster-whisper / whisper.cpp / Qwen3-ASR)、投影片換頁抓圖、OCR、一份可被機器檢查的筆記撰寫 規範;可當 Claude Code / Codex 技能使用,也可只當純 CLI 用。
狀態:v0.1.0.dev0 骨架。子命令介面、進度輸出、相依檢查與語言旗標已就位;各階段的實作 依序補上,尚未實作的階段會印 `[] not implemented yet` 並以 exit code 4 結束。
安裝
需要 Python 3.10 以上,以及 PATH 上的 `ffmpeg` / `ffprobe`。
python -m venv .venv
.venv\Scripts\pip install -e .[dev]
l2n --help
選用套件(extras):
| Extra | 用途 | 備註 |
|---|---|---|
breeze |
Breeze-ASR-25(預設引擎) | 需先跑 l2n convert-model 產生 CTranslate2 權重 |
qwen |
Qwen3-ASR 本機推論 | 開源權重;時間戳需另載 forced aligner,見引擎相容表 |
whispercpp |
whisper.cpp | 無 pip 相依,需自備編譯好的二進位並以 --whisper-cpp-bin 指定 |
scene |
PySceneDetect 場景偵測抓圖 | 未安裝時退回 ffmpeg scene filter |
dev |
pytest / pytest-cov | 開發與測試 |
轉錄引擎相容表
`l2n transcribe --list-engines` 會依本機現況印出同樣四行並標示相依是否滿足。
| 引擎 | local | needs_gpu | native_timestamps | 預設模型 | 相依 | 實測狀態 |
|---|---|---|---|---|---|---|
breeze_ct2(預設) |
yes | yes | yes | MediaTek-Research/Breeze-ASR-25 | faster-whisper + 自行轉檔的 CT2 權重 |
已在 Windows + RTX 4060 使用 |
faster_whisper |
yes | no | yes | large-v3 | faster-whisper |
CPU 可跑,有 CUDA 更快 |
whisper_cpp |
yes | no | yes | ggml-large-v3-turbo.bin | 自備二進位與 ggml 模型 | 以 --whisper-cpp-bin / --whisper-cpp-model 或 WHISPER_SRT_BIN / WHISPER_SRT_MODEL 指定 |
qwen3_asr |
yes | yes | yes(需對齊器) | Qwen/Qwen3-ASR-0.6B | pip install lecture2notes[qwen] |
待實測(見下) |
Qwen3-ASR 的三個實作前提
程式碼依 2026-09-21 查證的上游原始碼撰寫,以下三點與一般預期不同:
- ASR 模型本身不輸出時間戳。
Qwen3ASRModel.transcribe()只回language與text;cue 時間來自另一份第一方權重Qwen/Qwen3-ForcedAligner-0.6B(以forced_aligner=掛上,不需要外部 ctc-forced-aligner)。因此--model-dir之外另有--aligner-dir。 - 對齊結果是逐 token 的(中文為逐字),本套件在
cues_from_alignment()依句末 標點、靜默間隔、字數上限與時長上限重新組成 cue。時間戳單位是秒(上游型別標註 寫int,但回傳前已除以 1000)。 - 對齊器文件標示上限約 5 分鐘語音,所以本引擎宣告
chunk_sec = 240,轉錄階段會 自動切窗並把每窗的 cue 位移回整片的時間軸
Related Skills
Franken Whisper
Agent-first Rust ASR orchestration stack: Bayesian backend routing across whisper.cpp/insanely-fast-whisper/wh
Meeting Cleanup
Use when turning raw meeting notes, voice memos, or messy capture into a structured note with decisions, actio
Article Idea
Turn a raw topic or notes into a structured article brief — defines audience, angle, core argument, competitiv
Lemoncrow
A Complete faster runtime for coding agents. Optimize the journey, not the hop. Make coding agents 25% faster
Add Gemini Extension
Add a Gemini CLI extension as a Type 4 synced plugin to the marketplace. Automates the full workflow: meta.ts
Time Investment Report Agent Skill
An Agent Skill (Claude Code / WorkBuddy / CodeBuddy): before committing 10+ hours to a course, series, book or
Related Agents
Voice Engineer
GAIA voice interaction specialist. Use PROACTIVELY for Whisper ASR, Kokoro TTS, the Talk SDK, speech-to-speech
Editor Videos
Genera videos verticales para TikTok (edits de juegos/pelis y brain rots) con el pipeline local ffmpeg + Polli
Backend Engineer David
Use this agent when you need expert guidance on Rust systems programming, Tauri v2 desktop applications, audio