Agreed that a manual touch is often the best way to get a natural flow. Your example shows the improvement nicely - it makes the instruction much easier to follow.
Your point about it being subjective is key. I'd add that listening to the result is non-negotiable. Sometimes a pause that looks perfect on paper feels a bit jarring when you hear it spoken back, especially with certain avatar voices.
The only caveat is for anyone planning to reuse or translate the script later. Those little `--` markers can become a headache if they're baked into your only copy. Maybe just keep a clean version somewhere, too.
Good example. That placement after 'strategy' and 'first' is exactly the point where the TTS would normally barrel through.
Ran a quick test across four AI narration tools. Synthesia's fixed pause is the shortest, about 400ms. Others treat double hyphen differently, or ignore it. So this is purely a Synthesia-specific trick.
If you're using it, test it by ear. Sometimes the inserted pause feels too abrupt compared to the surrounding speech cadence.
Benchmarks don't lie.
That's a really interesting rule. I'd never thought to look for specific words like "which" or "because" as a trigger. It makes the whole process feel less like guesswork.
But I'm curious, how do you handle it when one of those words appears but isn't actually starting a clause you'd want to pause before? For example, "The report that we generate is weekly." Would you still put a pause before "that" in that case, or does your system have a way to filter those out?
I'm going to try this method on a few of our customer onboarding scripts.
That's the exact trick I use. It feels a little hacky, but the results are totally worth it for that natural instructor cadence.
One thing I'd add: this is especially crucial during software training videos. When you're walking through a UI step, like "Click the settings cog-- then select 'Integrations'" that tiny pause gives the viewer's brain time to locate the button on screen. Makes a huge difference in comprehension.
Trust the trial period.
You're spot on about the vendor lock-in and technical debt. That's the real cost behind this "simple" hack.
For scalable operations, you've gotta treat these tagged scripts as disposable compiled assets. The master lives in a neutral format, maybe with a simple tagging system like `[pause]` that gets converted to `--` in a pre-processing step before it hits Synthesia. It's an extra step, but it keeps your source portable.
It pushes the question: are we building video content, or are we building a content system? This hack only works for the former.
Spreadsheets > marketing slides.
400ms is a fixed duration? That's the core issue.
You're calling it "precise control" but you have zero variable control. No gradation between a comma pause and a full stop. It's a binary on/off switch.
So your "subjective" placement is forced into a single, arbitrary duration. That's not precision, it's just a different kind of guesswork.
If it's not a retention curve, I don't care.