How to Speak So Speech-to-Text Stays Clean

Speak in whole sentences at a normal pace

Speech models predict words from surrounding context, so the single biggest accuracy habit is giving them a full sentence to work with. Fragments strip the context that resolves ambiguous sounds, which is why dictating three words at a time produces worse results than dictating a paragraph.

• Decide the sentence before you start speaking it, then say all of it.
• Keep your conversational pace. Over-enunciation hurts more than it helps.
• Pause between sentences rather than in the middle of one.
• Keep volume steady to the end of the sentence, since trailing off costs the last few words.

This is a small mental shift with a large payoff. Instead of talking while you think, think for a beat and then talk. Two seconds of planning replaces ten seconds of editing, and the drafts come out sounding more like your writing because each sentence was finished before it was spoken.

Fix a wrong word without breaking the take

Everyone misspeaks. The habit that separates fast dictation from slow dictation is what you do next. Stopping the mic, deleting, and starting over costs a lot of time and breaks the thread of what you were saying.

Instead, correct in speech and keep moving:

• Say it plainly: we ship Tuesday, sorry, Wednesday.
• Restate the whole clause rather than a single word when the phrasing went sideways.
• Do not narrate meta commentary such as delete that, since it may end up in the text.
• Trust the cleanup pass and check the result once at the end.

Auri Voice resolves spoken self-corrections and false starts locally on the Mac, so this habit works with the network off. Practice it deliberately for a day. Once you stop treating a mistake as a reason to restart, dictation speed roughly doubles for most people, because the restarts were always the expensive part.

Punctuation: say it only where it matters

Older dictation required narrating every mark, and that habit lingers. Modern tools infer sentence boundaries well, so saying period after every sentence wastes effort and makes your delivery stilted.

Test your own tool with one paragraph before you commit to a habit. Dictate it once with no spoken punctuation and once with full narration, then compare the editing time. Most people find the first version needs less work, which frees attention for the content checks that actually require a human.

Names, numbers, and terms your model has never seen

Proper nouns are the errors that survive cleanup, because a wrong name is still a real word. These are worth handling deliberately, since they repeat in every message you write.

1. Add recurring names, products, and internal terms to a custom word list. Auri Voice supports custom words on Parakeet.
2. Say numbers in natural groups: fourteen fifty rather than one four five zero.
3. Speak dates the way you write them and check the weekday afterward.
4. For an unfamiliar name, say it once and fix it by hand rather than repeating it three times.
5. Type identifiers: email addresses, URLs, file paths, and codes.

The custom word list pays back more than anything else here. Five minutes adding twenty names and terms removes a category of correction you would otherwise make daily for months. Everything else on the list is a small habit; that one is a permanent fix, and it is the first thing to do after your microphone setup works.

Fix the room before you blame the model

Input quality bounds everything downstream. A close, consistent microphone in a quiet-enough room does more for accuracy than any engine change, and it costs nothing.

• Get the mic six to twelve inches from your mouth, off to the side rather than directly in front.
• Pick one input device and stop switching between AirPods, laptop mic, and a webcam.
• Turn off steady noise you control: a desk fan, a nearby speaker, a rattling vent.
• Speak across the mic, not into it, so plosives do not clip.
• In a cafe, turn away from the loudest source and accept shorter takes.

If you dictate on the move, a close mic matters even more. The Apple Watch works well here because your wrist sits inches from your mouth, and with Auri Voice that audio is encrypted in transit and transcribed locally on the Mac. Distance beats model size in noisy places, every time.

A five-minute practice script

Habits form faster with deliberate practice than with hoping. Spend five minutes once, in Notes, where nothing you say matters.

1. Dictate a three-sentence paragraph about your day with no spoken punctuation.
2. Dictate the same paragraph and deliberately misspeak once, correcting in speech.
3. Dictate a sentence with two colleague names, a price, and a date.
4. Dictate a four-item list, saying comma between items.
5. Read all four results and note which errors survived.

The surviving errors are your real task list. If names failed, build the word list. If the last words of sentences vanished, work on volume and breath. If punctuation was wrong inside lists, adjust how much you narrate. Re-run the script after a week and you will see which habits stuck, which is more useful than judging a tool by a single frustrating afternoon.

FAQ