Voice recognition not working — what to do?

Check microphone permission, network, and language settings. Most issues are permission-related and resolve in under 60 seconds.

Voice capture failing is the most common help request because it has many possible causes — permissions, network, microphone hardware, language settings, accent and pronunciation handling. The fix is usually one specific thing rather than something complex; this article walks through the diagnostic order from most likely to least.

First: check microphone permission

About 70% of voice-capture failures we see are permission-related. The app needs microphone access; if iOS or Android revoked it (which can happen after OS updates), no audio reaches the server.

On iOS:

Open Settings (the gear app, not in KeptMind)
Scroll down and tap KeptMind
Ensure Microphone is on
Also enable Speech Recognition if listed
Reopen KeptMind and test capture

On Android:

Open Settings → Apps → KeptMind → Permissions
Ensure Microphone is allowed
On Android 13+, ensure Background allowed if you use voice from the lock-screen widget
Reopen and test

Second: check network

Voice transcription requires uploading audio to our servers. If you are on a slow or unstable connection (poor mobile signal, captive Wi-Fi like a coffee-shop or hotel network), the upload may time out and the capture fails silently.

Test by switching to a different network — connect to your home Wi-Fi if you were on cellular, or vice versa. If voice works on one network and not another, the issue is network-side, not app-side. Captive portals (login screens at hotels, airports, cafes) frequently block audio upload until you complete sign-in; try opening any web page in your browser to trigger the captive portal flow first.

Third: try a quiet environment

Background noise reduces transcription accuracy substantially. If captures are happening but coming back as gibberish, try the same phrase in a quiet room. ADHD speech under noise (kitchen, car, walking outside) often transcribes to nothing because the speech-to-text engine cannot separate your voice from ambient sound.

For consistently noisy environments, consider:

Bringing the phone closer to your mouth (under 30cm reduces noise pickup substantially)
Using AirPods or Bluetooth headset with built-in microphone — the microphone proximity helps the engine focus on your voice
Switching to text capture for those moments — typing is reliable in noise where voice is not

Fourth: check language setting

KeptMind supports voice transcription in English (best accuracy) and Estonian (good accuracy on standard speech). Other languages work via Google's speech engine but accuracy varies.

If your captures are being transcribed in the wrong language (e.g. you speak English but the transcript comes out as garbled Estonian-looking text), check Settings → Voice → Capture language. Set it to your primary speaking language. The language setting affects how the engine interprets phonemes; mismatched setting produces consistently garbled output.

Fifth: check microphone hardware

Less common but worth ruling out — your phone's microphone itself may be physically blocked or damaged. Test by recording a voice memo in your phone's built-in voice recorder app (iOS Voice Memos, Android Recorder). If that fails too, the issue is hardware, not KeptMind. Common physical issues: dust in the mic port, screen protector blocking the mic, or water damage from a recent splash.

When the standard fixes do not work

A small fraction of voice-capture issues come from edge cases:

Custom accessibility settings — VoiceOver, TalkBack, or other assistive tech can interfere with the capture flow. Try toggling these temporarily to test.
VPNs — some VPN configurations route audio through unstable paths. Disable VPN, test capture, then re-enable.
Older phones — devices older than 5 years may struggle with the audio upload because of slower processors. Capture still works but takes longer; if it consistently times out, this may be the cause.
Specific Android skins — some manufacturer skins (especially older Samsung One UI versions) have aggressive battery optimization that can kill the capture mid-upload. Whitelist KeptMind in battery optimization settings.

When to email us

If you have tried the steps above and voice still fails, email hello@keptmind.com with:

Device model (e.g. iPhone 14 Pro, Samsung Galaxy S23)
OS version (Settings → General → About on iOS; Settings → About phone on Android)
KeptMind version (Settings → About → Version)
A description of what happens (silent fail, garbled text, app crash, etc.)
Whether the built-in phone voice recorder works

We respond within 1 business day for voice-capture issues because we know it blocks daily use. About 90% of cases resolve via email troubleshooting; the remaining 10% are real bugs we fix in the next release.

Frequently asked questions

Why does voice work on iOS but not Android (or vice versa)?

iOS and Android use different underlying speech engines (Apple Speech vs Google Speech). The audio path also differs slightly — iOS captures via AVAudioEngine, Android via AudioRecord. Bugs occasionally affect one platform and not the other; if you see a platform-specific issue, that is normal and we treat platform-specific bug reports with priority.

Can I use KeptMind voice capture on a smartwatch?

Apple Watch capture is in beta — it works but feels rough. Wear OS support is on the roadmap, not yet shipping. For now, we recommend phone-based capture for reliability.

How accurate should I expect transcription to be?

For clean English speech in a quiet environment: 95-98% word accuracy. For ADHD-style speech (fast, hesitant, mid-stride) in normal environments: 85-92%. For non-English or accented English in noisy environments: 75-85%. The captured task usually remains recognizable even at 75% accuracy because the AI parsing tolerates common transcription errors. If your accuracy is consistently below 80%, the workarounds above (language setting, environment, custom vocabulary) usually help.

Are there other voice tips?

Three small habits that consistently help: speak the date naturally ("Friday at 3pm" works better than "tomorrow"); name the person directly ("text Marek about budget" beats "text him about it"); and use action verbs first ("schedule dentist" beats "I should probably schedule the dentist sometime"). The AI parsing keys on these patterns and produces cleaner tasks.

For a deeper dive into how voice capture is supposed to work in normal use, see the voice capture article. For an SEO/research view of the broader voice-first ADHD productivity question, see the blog post on voice-to-task for ADHD.