Training Lecture Transcript

---
name: training-lecture-transcript
sheetId: "4.01"
description: "Polish long video lecture transcripts into lecturer-friendly readout scripts in Traditional Chinese."
category: training
project: training-lecture-transcript
platform: multi
status: Done
author: Peter Tu
input: "Raw 30-90 minute lecture transcript text with filler words and STT artifacts"
process: "Sequential batch processing (~700 Chinese chars per batch) with strict order gating, filler removal, and sentence restructuring"
output: "Lecturer-friendly readout script in Traditional Chinese with terminology consistency notes"
synergy: []
house-style: exempt — verbatim capture — fidelity to the source outranks house voice

---

# Polish Lecture Transcript

```bash
npx skills add https://github.com/peter-tu-zynkr/zynkr-skill-builder --skill training-lecture-transcript
```

Process long transcript text into lecturer-friendly readout content. Use this skill when you've recorded a 30–90 minute Chinese lecture and want a clean Traditional Chinese readout script — filler words removed, STT artifacts fixed, run-on speech broken into short readable sentences, all without losing terminology or speaker intent.

## Workflow

1. Normalize input text.
2. Split transcript into ordered batches of ~700 Chinese characters.
3. Process exactly one batch at a time.
4. Do not start batch N+1 until batch N is completed and checked.
5. Merge completed batches and run final consistency pass.

## Batch Gate Rules

- Maintain strict order: `001 -> 002 -> 003 ...`.
- Keep a visible status for each batch: `PENDING | IN_PROGRESS | DONE`.
- Move to the next batch only when current batch is `DONE`.
- If text is ambiguous, add a short editor note in the current batch; do not skip ahead.

## Per-Batch Rewrite Objective

For each batch, keep meaning and remove speech noise.

### Must do

- Remove filler/disfluency words (see `./references/filler_words_zh.md`).
- Fix obvious STT errors and repeated fragments.
- Convert run-on speech into short, readable sentences.
- Preserve terminology, examples, and speaker intent.
- Keep original language (Traditional Chinese by default).

### Must not do

- Add new claims not present in the source batch.
- Reorder major logic across batches.
- Mix two batches in one rewrite.
- Jump to later batches before current batch is done.

## Output Format

Use this structure for each completed batch:

```markdown
### Batch 001
- Status: DONE
- Source length: <N chars>
- Edited length: <N chars>

<rewritten text>
```

After all batches are done, produce:

1. `Merged Script (full)`
2. `Terminology/consistency notes` (only if needed)

## Recommended Command Helpers

Use bundled scripts for deterministic batching:

- Split text into batches:
  - `python3 scripts/chunk_transcript_zh.py <input.txt> --max-chars 700 --out batches.md`
- Validate order and status gate:
  - `python3 scripts/chunk_transcript_zh.py --validate batches.md`

## Readout Style Defaults

- Sentence length: short and easy to read aloud.
- Tone: lecturer-friendly, concise, direct.
- Paragraphing: split by idea shifts, not by arbitrary line breaks.
- Keep key terms in English when they are standard in context (e.g., IPO, SOP, RAG), otherwise use Traditional Chinese.

## Optional Final Transform

If user asks for chapterized output, reorganize only after all batch rewrites are done.