Agents365-ai/video-podcast-maker / skills/video-podcast-maker-lite
video-podcast-maker-lite
VerifiedMinimal personal narrated-video pipeline — a topic becomes a talking-head-free explainer MP4 (1080p or 4K) via script → Azure TTS (SSML) → Remotion. Use when the user wants a quick narrated video from a topic without the full video-podcast-maker machinery (no extra skills, no thumbnails/shorts/publish matrix). Do NOT trigger for heavy production needs — use video-podcast-maker for those.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 18.4 KB |
| references/pronunciation.md | markdown | 3.0 KB |
| scripts/tts.py | script | 27.0 KB |
references/pronunciation.md
View on GitHub# Pronunciation reference (zh-CN) — video-podcast-maker-lite
Loaded from SKILL.md Step 2. Everything here is about making Azure TTS say
Chinese tech content correctly. The spoken layer in `tts.py` does the mechanical
work; the author's job is to feed the dicts.
## Polyphone pre-flight (before TTS)
Scan the script for polyphone words — 命令行/一行 (háng), 同行 (háng), 重做/重新
(chóng), 增长 (zhǎng), 银行 (háng), etc. The global dict
`~/.video-podcast-maker/phonemes.json` already covers common tech terms — check
it first; add missing ones as **whole-word** entries (never single characters)
to `videos/{name}/phonemes.json`. `tts.py` turns them into SSML `<phoneme>`
tags automatically — don't wait for the preview listen to catch a misreading.
`tts.py` runs a polyphone pre-flight on every invocation; `--check` makes it
lint-only (no synthesis): it flags minority-reading shapes (计数+行 háng,
重做/重灌… chóng, 长会话/长上下文 cháng, …) that no phoneme-dict entry covers —
fix the dict before synthesizing, don't discover misreadings at the preview
listen.
## Known misreading shapes (auto-fixed by the spoken layer)
All of these are handled by `tts.py` automatically, except brand names — add
those to `aliases.json` yourself:
- Decimals that should read digit-by-digit after the point: `1.5` → 一点五,
not English "one point five".
- Letter-by-letter acronyms: `MoE` → M-O-E, not "莫".
- Letter+digit combos: `FP8` → F-P-八.
- Model-name runs that TTS breaks wrongly: `9B Dense、35B MoE、397B MoE` —
digits + 顿号 break cleanly; spelled-out `九B 稠密、三十五B MoE` mis-breaks
into things like "9BMoe".
## Shapes that always need `aliases.json` entries
- Hyphenated model ids: `deepseek-v4-flash-vision-exp` → `DeepSeek V four
Flash Vision E X P`.
- camelCase benchmark names: `DeepSWE` → `Deep S W E`.
- Code identifiers: `file_id` → `file I D`, `base64` → `base 六十四`.
Aliases may themselves contain hyphens — `pronounce()` matches them with
`startswith` at ASCII-letter positions, so a hyphenated alias still hits.
## Alias dict mechanics
Brand-name readings live in `aliases.json` (display → spoken, e.g.
`"Qwen": "千问"`):
- per-video entries: `videos/{name}/aliases.json`
- channel-wide: `~/.video-podcast-maker/aliases.json`
- merged over the built-ins in `tts.py` (`PRONUNCIATION_ALIASES` /
`LETTER_WORD_ALIASES`)
## Built-in spoken-layer behaviors
- A spoken number followed by a single-Hanzi quantifier is glued into one unit
(`10 倍` → 十倍, `57 分` → 五十七分) — otherwise Azure inserts a pause
between the number word and the quantifier; extend `_CN_QUANTIFIERS` in
`tts.py` if a new quantifier stumbles.
- `1M` is built in as 一兆 (so `1M token` reads "一兆", not letter M; other SI
suffixes like `T` stay letter readings).
## Why not SSML `<sub alias>`
Tried and abandoned: Azure's word-boundary events for `<sub>` are buggy and
corrupt the SRT. The spoken layer + boundary mapping in `tts.py` is the
replacement.Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| writing-skillsobra/superpowersUse when creating new skills, editing existing skills, or verifying skills work before deployment | Stars: 323k | Updated | Claude CodeCodex | Verified | Install |
| xlsxanthropics/skillsUse this skill any time a spreadsheet file is the primary input or output. This means any task where the user wants to: open, read, edit, or fix an existing .xlsx, .xlsm, .xltx, .csv, or .tsv file (e.g., adding columns, computing formulas, formatting, charting, cleaning messy data); create a new spreadsheet from scratch or from other data sources; or convert between tabular file formats. Trigger especially when the user references a spreadsheet file by name or path — even casually (like "the xlsx in my downloads") — and wants something done to it or produced from it. Also trigger for cleaning or restructuring messy tabular data files (malformed rows, misplaced headers, junk data) into proper spreadsheets. The deliverable must be a spreadsheet file. Do NOT trigger when the primary deliverable is a Word document, HTML report, standalone Python script, database pipeline, or Google Sheets API integration, even if tabular data is involved. | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| shipping-and-launchaddyosmani/agent-skillsPrepares production launches. Use when preparing to deploy to production, or when asking what needs to be in place before shipping. Use when you need a pre-launch checklist, when setting up monitoring, when planning a staged rollout, or when you need a rollback strategy. | Stars: 119k | Updated | Claude CodeCodex | Verified | Install |
| childhoodthedotmack/claude-memBabysit a shipped release after merge, publication, or deployment. Verify published packages, production behavior, installed plugins, and late CI or user reports until the observation window is clean; use babysit for PR readiness before release. | Stars: 118k | Updated | Claude CodeCodex | Verified | Install |