Lecture Notes

Before using audio or video transcription, make sure OPENROUTER_API_KEY is defined in the environment.

Use OpenRouter's chat completions API with multimodal input_audio. The prompt should explicitly ask for a verbatim transcript with minimal cleanup.

AUDIO_BASE64=$(base64 < "Lecture 05.audio.wav" | tr -d '\n')

curl https://openrouter.ai/api/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $OPENROUTER_API_KEY" \
  -d "{\n    \"model\": \"openai/gpt-4o-audio-mini\",\n    \"messages\": [\n      {\n        \"role\": \"user\",\n        \"content\": [\n          {\n            \"type\": \"text\",\n            \"text\": \"Transcribe this lecture audio as plain text. Keep the wording faithful to the speaker, preserve important structure, and do not summarize.\"\n          },\n          {\n            \"type\": \"input_audio\",\n            \"input_audio\": {\n              \"data\": \"$AUDIO_BASE64\",\n              \"format\": \"wav\"\n            }\n          }\n        ]\n      }\n    ]\n  }"

For actual skill usage, save the JSON response first, then extract the transcript into a plain text file:

AUDIO_BASE64=$(base64 < "Lecture 05.audio.wav" | tr -d '\n')

curl https://openrouter.ai/api/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $OPENROUTER_API_KEY" \
  -d "{\n    \"model\": \"openai/gpt-4o-audio-mini\",\n    \"messages\": [\n      {\n        \"role\": \"user\",\n        \"content\": [\n          {\n            \"type\": \"text\",\n            \"text\": \"Transcribe this lecture audio as plain text. Keep the wording faithful to the speaker, preserve important structure, and do not summarize. Return only the transcript text.\"\n          },\n          {\n            \"type\": \"input_audio\",\n            \"input_audio\": {\n              \"data\": \"$AUDIO_BASE64\",\n              \"format\": \"wav\"\n            }\n          }\n        ]\n      }\n    ]\n  }" \
  > "Lecture 05.transcript.response.json"

jq -r '.choices[0].message.content // (.choices[0].message.content[]? | select(.type == "text") | .text)' \
  "Lecture 05.transcript.response.json" > "Lecture 05.transcript.txt"

Recommended Local File Outputs

Given an input file like Lecture 05.mp4 or Lecture 05.m4a, save derived files alongside it using predictable names:

• Lecture 05.audio.wav for audio extracted from video or normalized for upload
• Lecture 05.transcript.response.json for the OpenRouter API response
• Lecture 05.transcript.txt for the plain text transcript used to generate notes

Video to Audio Extraction

Use ffmpeg to extract audio from video before transcription. A safe default is:

ffmpeg -i "Lecture 05.mp4" -vn -ac 1 -ar 16000 -c:a pcm_s16le "Lecture 05.audio.wav"

If the source is already audio but not in a convenient format, normalize it to .wav before base64 encoding and upload.

Transcript Handling Rules

• Always generate notes from the saved local transcript file, not directly from the raw media file.
• If the user also provides slides, PDFs, or other lecture materials, combine them with the transcript when producing notes.
• Preserve important spoken structure such as section transitions, definitions, theorem statements, examples, and repeated exam hints.
• If the transcript is clearly noisy or incomplete, note uncertainty in the final notes rather than silently inventing missing content.
• If a transcript already exists locally for the same media file, reuse it unless the user explicitly asks for a fresh transcription.

Output Format

Output all notes inside fenced codeblocks. When content spans multiple chapters or distinct topics, create separate files.

File Naming

Chapter X - Title.md or Unit X - Title.md

Document Structure

---