Voxor
Download 
You are partway through a sentence, you stop to think of the right word, and by the time you have it the microphone is off. Everything after that went nowhere.
If you have hit this, you have probably also read that Mac dictation has a thirty second limit. That is close enough to be useful and wrong in the way that matters.
What Apple actually documents
From Apple’s own Mac User Guide, in their words:
Dictation stops automatically when no speech is detected for 30 seconds.
Read it twice, because the difference between that sentence and the popular version is the whole article.
The popular version is a cap on talking. Thirty seconds of speech and you are cut off, like a meter running out. If that were true, the fix would be obvious: talk faster, or talk in shorter bursts.
What is documented is a timeout on silence. The clock is not measuring how long you have been speaking. It is measuring how long you have not been.
Why that is worse, not better
A cap on talking would be annoying and predictable. You would learn its length and work inside it.
A timeout on silence is neither, because silence is not idleness. The pauses in real dictation are the load-bearing parts of it. You stop to choose between two words. You stop to check what you just said. You stop because someone asked you something, or because you reread the paragraph above, or because the thought was genuinely hard and you needed a second with it.
None of that is inactivity. All of it counts as no speech detected.
So the feature is not cutting you off for going on too long. It is cutting you off for thinking, and it does so at exactly the moment thinking was going well. That is why it feels personal.
It also explains a pattern people report and then doubt themselves about: dictation works fine when you are reading something aloud, and falls apart when you are composing. Reading has no gaps. Composing is mostly gaps.
Why starting it again does not fix it
The advice you will find is to press the key again and carry on. It works, in the sense that the microphone comes back.
What it does not restore is the sentence you were in the middle of. The cost of the cutoff was never the seconds; it was that you were holding a thought, and the interruption made you drop it and restart it from the top. Anyone who has tried to dictate a long email knows the second attempt is worse than the first.
A tool that interrupts you every time you pause is not a slow tool. It is a tool you cannot think in.
What continuous dictation asks for instead
A dictation tool that never stopped on its own would be worse, not better: a microphone left open by a forgotten keystroke is its own problem. So the question is not whether there is a silence rule. It is how much room it leaves you, whether it warns you, and what happens to what you already said.
Voxor has a silence rule too, and here it is precisely. Sixty seconds of continuous quiet ends a session, with an on-screen warning at twenty, and the clock resets every time you speak again, so it measures the current gap rather than the whole session. Two things separate that from being cut off. You get told before it happens rather than discovering it afterwards, in those words:
No speech detected. Sending in 40s.
And when it does stop, it submits what you said instead of discarding it. The warning tells you both facts at once, which is the whole difference between a tool stopping and a tool abandoning you.
That is a difference of degree on the timer and a difference of kind on the outcome. Thirty silent seconds is inside the length of an ordinary pause for a word. Sixty, announced at twenty, is long enough to think in.
The second half matters as much. Speech that survives a long pause tends to arrive messy: a false start, a correction halfway through, a sentence you abandoned and rebuilt. The formatting layer is what turns that into something you would send, and it takes account of where the text is going, so a terminal gets literal characters and a message gets ordinary punctuation.
The honest boundaries
Four, because a claim with the edges sanded off is not much of a claim.
This is about the built-in feature, not about Apple. The thirty second rule is documented, and a timeout is a reasonable default for a system feature that anyone might trigger by accident. It is a poor fit for someone dictating all day, which is a different job.
Behaviour changes across versions and settings. We have quoted the current documented rule. If your Mac behaves differently, believe your Mac.
Voxor runs on Mac and Windows today. iPhone and Android are in closed beta and not yet publicly available, so none of this describes a phone.
Local Mode has a hardware floor. Running recognition and formatting on your own machine needs real headroom, and on a very constrained one Voxor declines rather than pretending.
The short version
Mac dictation does not limit how long you can speak. It limits how long you can go without speaking, and thirty seconds is not very long when you are trying to say something you have not fully worked out yet.
If the pauses are the problem, what you want is not a tool with no rule at all. It is one whose rule leaves room for thinking, tells you before it acts, and keeps what you already said. Start with Voxor for Mac, and if you want to know what leaves your machine and what does not, we set that out in what Voxor sends to the cloud.