Voice Typing Keyboard: What to Look For

Start from the messages you send this week

Feature lists are written to look complete. Your week is not. Before you compare anything, write down the five kinds of messages you actually send, because that list is the only scorecard that predicts whether a keyboard survives past day three.

A typical list looks like this:

• Short chat replies where speed matters more than polish.
• Work emails that need to read like you wrote them carefully.
• Long explanations to a teammate or a client.
• Messages in a second language, or with names from one.
• Notes to yourself that nobody else will read.

Now rank those by volume. If four of five are short chat replies, mic speed and switching friction outweigh a large rewrite panel. If work email dominates, judge the reply and grammar tools hardest. This step takes two minutes and it stops you from picking a keyboard that wins on the features you will never press.

Microphone behavior worth testing

Recognition accuracy is one part of the mic experience, and the parts people quit over are timing and recovery. A keyboard that cuts you off mid-thought is annoying in a way that a slightly worse transcript is not.

Test these five behaviors with the same spoken paragraph:

1. Start latency: how long between tapping the mic and being able to speak.
2. Pause tolerance: whether a two-second thinking pause ends the take.
3. Long-form stamina: whether a ninety-second paragraph survives intact.
4. Noise handling: same paragraph near a fan, on a street, or in a cafe.
5. Recovery: what happens when you correct yourself out loud.

Score the last one carefully. Everyone misspeaks, and the difference between a keyboard that leaves both versions in the text and one that resolves the correction is a real time saving on every message. Repeat the noisy test, because ambient sound is where fragile demos fall apart and where you will spend a good share of your dictation life.

What happens after the words land

Raw transcription is the easy half. The value shows up in what the keyboard does with the text once it exists, since that determines whether you send the draft or rewrite it by thumb.

Test each tool on the same dictated paragraph rather than on a clean sample sentence. Cleanup tools that shine on tidy input and mangle real speech are common, and the only way to see that is to feed them the messy paragraph you would actually produce while walking to a meeting.

Permission honesty as a scoring row

Every keyboard with cloud features needs Full Access, because iOS blocks network calls from keyboard extensions without it. Every candidate will ask for it, so score them on how plainly each one explains what it does with the access.

Look for these specifics in the app's own words:

• Which features require the network and which work locally.
• What is sent for those features: audio, text, or both.
• Whether transcription can run locally as an option.
• A clear statement about password and card fields.
• Whether typing still works if you decline or revoke Full Access.

Auri Keyboard uses cloud transcription by default on iPhone and offers local transcription, requires Full Access for AI features, and does not capture password or card fields, which iOS blocks for all custom keyboards anyway. Vague privacy adjectives with no per-feature detail should cost an app points. Specific answers, even ones you dislike, are the sign of documentation someone actually maintains.

Switching friction decides whether you keep it

An excellent keyboard you forget to switch to is worth nothing. The globe key press is the tax you pay on every message, and small details around it decide whether the habit forms.

• Keep only the keyboards you use, so the globe list is short.
• Check whether the app remembers your choice per app or resets constantly.
• Time how long the keyboard takes to load in a cold app. Half a second feels different from two.
• Confirm the mic button sits where your thumb already rests.
• Watch for layout shifts between the stock keyboard and this one, since muscle memory hates surprises.

Run this check on the phone you actually carry, not a newer one. Keyboard extensions get evicted from memory more aggressively on older hardware, which turns a snappy demo into a pause every time you switch apps. If that pause lands in the middle of a fast conversation, you will drift back to the stock keyboard without deciding to.

A trial protocol you can finish

Give each candidate a week with fixed material, and compare notes at the end rather than impressions along the way. Impressions favor whichever app you tried most recently.

1. Write three test messages: a chat reply, a work email, and a long explanation.
2. Install one keyboard and use it as your default for a week.
3. Dictate the three test messages on day one, day four, and day seven.
4. Track two numbers: taps needed to fix each draft, and times you gave up and switched away.
5. Repeat with the second candidate, then compare the numbers.

Keep Apple Dictation enabled throughout so you always have a fallback and a baseline. The give-up count is the most honest metric in the set, because it captures friction that no feature table shows. A keyboard you abandoned twice in a week during a real conversation is not going to win in month two, whatever its rewrite panel offers.

FAQ