mirror of
https://github.com/heygen-com/hyperframes.git
synced 2026-09-14 18:01:20 +08:00
ukimsanov
71 lines
2.5 KiB
Markdown
71 lines
2.5 KiB
Markdown
# SCRIPT — example.com intro
|
||
|
||
**Voice:** River — ElevenLabs, voice id `SAz9YHcvj6GT2YYXdXww`
|
||
**Model:** `eleven_multilingual_v2` · `mp3_44100_128` → 44.1 kHz mono WAV
|
||
**Voice direction:** Calm and unhurried. Low-key confidence, no sell. Read it the way
|
||
you would read a footnote you happen to find interesting.
|
||
|
||
The whole script is **one** synthesis call so the sentence-to-sentence prosody is real
|
||
rather than three clips butted together. The `**Time:**` values below are not
|
||
estimates — they are the measured word timings from
|
||
`npx hyperframes transcribe assets/narration.wav --model small.en`, offset by the
|
||
narration clip's `data-start` of `1.00s`.
|
||
|
||
---
|
||
|
||
## Line 1 — Name it (Frame 1)
|
||
|
||
**Time:** 1.10 – 2.81s
|
||
**Delivery:** Flat, factual. The period is real; let it land.
|
||
|
||
This is example dot com.
|
||
|
||
## Line 2 — What it is for (Frame 1)
|
||
|
||
**Time:** 2.81 – 5.38s
|
||
**Delivery:** Slightly warmer. "Reserved" carries the sentence.
|
||
|
||
The domain reserved for documentation examples.
|
||
|
||
## Line 3 — The payoff (Frame 1)
|
||
|
||
**Time:** 5.48 – 7.68s
|
||
**Delivery:** The lift of the piece, but quiet. Space before "no permission needed."
|
||
|
||
Use it in your docs — no permission needed.
|
||
|
||
---
|
||
|
||
## Spoken vs. captioned
|
||
|
||
One deliberate divergence: the TTS text says **"example dot com"** because that is how
|
||
the string is pronounced; the caption reads **"example.com"** because that is how the
|
||
string is written. Whisper transcribed the spoken form back as `example.com,`, which is
|
||
why the caption group's word timing for it spans a full 1.27s.
|
||
|
||
## Regenerating
|
||
|
||
```bash
|
||
# 1 — synthesize (ElevenLabs, via the /media-use audio engine)
|
||
node -e '
|
||
import("'"$HOME"'/.claude/skills/media-use/audio/scripts/lib/tts.mjs").then(m =>
|
||
m.synthesizeOne({
|
||
provider: "elevenlabs",
|
||
voiceId: "SAz9YHcvj6GT2YYXdXww",
|
||
text: "This is example dot com. The domain reserved for documentation examples. Use it in your docs — no permission needed.",
|
||
wavAbs: process.cwd() + "/.work/narration.raw",
|
||
hyperframesDir: process.cwd(),
|
||
}).then(r => console.log(r)))'
|
||
|
||
# 2 — normalize to 44.1k mono PCM
|
||
ffmpeg -y -i .work/narration.raw -ac 1 -ar 44100 -c:a pcm_s16le assets/narration.wav
|
||
|
||
# 3 — word timings for the caption groups
|
||
npx hyperframes transcribe assets/narration.wav --model small.en
|
||
```
|
||
|
||
Step 3 rewrites `.work/transcript.json`. If the timings move, the caption group
|
||
boundaries in `compositions/captions.html` and the `WORDS` table in this file's
|
||
consumer must move with them — they are hand-committed on purpose so the composition
|
||
has no build step.
|