How to choose speech to text software by testing it on your own voice

There are two kinds of speech to text software. The first turns your voice into words, close to what you said, fillers and false starts included. The second turns your voice into writing you can send: it removes the ums, applies your corrections and adds punctuation and paragraphs. If your text goes only into your own notes, the first kind may be enough. If other people read it, the second kind saves you the cleanup.

Word accuracy is no longer the main difference between tools. Most modern models hear clear speech well. What separates them is what happens after the words are recognized.

• Cleanup. Does it remove fillers, repetitions and false starts, and apply corrections such as "no, make that Thursday"?
• Formatting. Does it add punctuation, paragraphs and lists without spoken commands for each comma?
• Styles. Can a chat reply come out short and casual while a client email comes out complete and polite?
• Where it works. Native desktop app, browser, extension or phone. It has to work where you type, not where the vendor prefers.
• Vocabulary. Is there a personal dictionary for names, products and jargon, so they come out right the first time?
• Editing by voice. Can you change the last sentence or make a paragraph shorter without reaching for the keyboard?
• Languages. Does it handle the languages you write in, and switching between them?
• Privacy. Is audio deleted after processing? Is your text used to train models? Can you delete your history?
• Price and limits. What it costs monthly and yearly, and how usage is measured, by minutes, words or not at all.

1. 01 Where do you write most: in the browser, in desktop apps or on your phone?
2. 02 Do other people read what you dictate, or is it notes for yourself?
3. 03 Roughly how many minutes a day would you dictate?
4. 04 Which names and terms come up often in your work?
5. 05 Do you work alone, or does a team need shared names and styles?

Your answers narrow the list faster than any comparison table. Write them down, because you will use them in the test below.

Pick no more than three tools to test. More than that and the test turns into a chore you will not finish. A good short list usually has one built-in option, because it costs nothing extra and sets the baseline, and two AI dictation apps that work where you write.

Built-in tools such as Apple Dictation, Windows Voice Typing with Win+H, Google Docs Voice typing and Gboard are fine for short notes and quick replies. They usually leave fillers and corrections in and have no styles, which is exactly what the test will show. See our guides to voice typing in Google Docs and Windows voice typing.

If you want a side by side view of the main tools with prices at the time of writing, read our comparison of the best speech to text software. This guide is about how to choose, that page is about what is on the market.

Run the same test in each tool on your short list. It takes about twenty minutes per tool.

1. 01 Pick one real task you do every day, such as a reply to a client email.
2. 02 Dictate it the way you would explain it to a colleague, with your normal pauses, a correction or two and one name that is hard to spell.
3. 03 Count the edits you need before you would send the result. A changed word is one edit, a fixed sentence is one edit.
4. 04 Repeat it in the three apps you write in most, for example mail, chat and a document.
5. 05 Dictate one longer piece of about two minutes, such as meeting notes, and check the paragraphs.
6. 06 Try one edit by voice, such as making the last sentence shorter.

Write the edit counts in a small table. The tool with the fewest edits in your real apps is the right one. A long feature list does not help if you still retype half the result.

• Reading a prepared text out loud. Nobody dictates like that, and every tool looks good on clean input.
• Testing in a quiet room only. Try it once with normal background noise from your work day.
• Judging on the first sentence. Give each tool a few minutes, since your speaking rhythm settles quickly.
• Skipping the names. If the tool misspells your clients and products, you will fix them every single day.

Dictation means sending your voice somewhere unless the model runs on your device. Read the privacy page of each tool on your short list and look for clear answers on three points: is audio deleted after transcription, is your data used to train models, and can you switch history off or delete it. If a vendor does not answer these plainly, treat that as an answer.

VoiceTyping deletes audio right after transcription, never uses your audio or text to train models, encrypts text history at rest and lets you switch history off or delete everything in one click. Details are on the privacy and security page.

Compare cost by the amount you dictate, not by the headline number. Work out your minutes a day from the test, multiply by your working days and check whether the plan covers it with room to spare. Look at the yearly price too, since some tools discount yearly billing far more than others.

VoiceTyping plans have clear AI minute budgets: Personal 400 minutes a month for 15 USD monthly or 7 USD per month billed yearly, Pro 900 minutes for 29 USD or 14 USD, and Team 1,500 minutes per seat for 49 USD or 24 USD per seat, from 3 seats. The yearly price saves 50%. See the voice typing app pricing page for every detail.

VoiceTyping runs in the browser, as an installed web app and as an extension for Chrome and Edge that types into text fields on the web. It removes fillers, applies your corrections, adds structure and picks a style by the site you are on. It has no native desktop app and no system-wide hotkey outside the browser, so it suits people who write mostly in browser tools such as Gmail, Outlook on the web, Slack, Notion and Google Docs. If you spend the day in desktop programs, include a tool with native apps on your short list.