Switching Mac Dictation Apps: A Migration and Test Checklist

Save vocabulary and replacements, check import compatibility, set a conflict-free shortcut, and test text quality, clipboard, terminals and offline fallback.

Source and cited vendor documentation reviewed October 17, 2026 against the 0.3.7 build 42 checkout. Public v0.3.7 is build 35; controls may differ. The sample sentences below are an acceptance-test protocol, not a cross-app benchmark. Check the public release

Quick answer

Keep the old app available while you save vocabulary, replacements, shortcuts and billing details. Ducduc imports its own dictionary backup, not arbitrary competitor exports. Configure permissions, speech assets and cleanup routing, then compare raw text, final text and insertion in your real apps before completing the switch.

The safest way to switch dictation apps is to keep the old setup available, move a small useful vocabulary, and test the new app in the places where you actually type. A successful demo in Notes does not establish that your terminal, bilingual terms or clipboard will behave the same way.

This checklist works whether you are leaving Wispr Flow, superwhisper, Typeless or another tool. The Ducduc-specific steps are based on the reviewed source build; they do not promise automatic migration from those products.

Before uninstalling, save the parts you would miss

Keep a readable copy of these items while the old app still opens:

ItemWhat to saveHow it maps to Ducduc
VocabularyNames, products and jargon with exact spellingPersonal Dictionary, one term at a time
ReplacementsWrong phrase → intended phrase, including app scopeExact replacements; review each rule
Snippets or promptsTrigger and complete saved textRecreate deliberately; not part of dictionary import
Workflow settingsKey, tap/hold gesture, language, microphone, cleanup styleReconfigure in Settings
History you needRelevant text through the old app’s supported methodNo competitor-History importer is promised
BillingRenewal date, purchase channel and license informationSeparate from copying app data

Use the old app’s supported export if available; otherwise save a readable manual list. Do not assume that a feature called “Import” also offers an exporter.

Formats differ. Wispr Flow documents CSV dictionary imports and JSON snippet imports. VoiceInk documents a JSON vocabulary/replacement archive. Neither format is automatically a Ducduc backup.

Ducduc’s Import… expects its own dictionary-backup structure. Renaming a competitor file to .json does not convert it. There is no supported generic CSV or bulk-paste importer in the reviewed Dictionary UI. If you have hundreds of terms, allow time for manual entry rather than assuming a one-click move.

Save the old app’s settings, then quit it fully and pause its automatic startup while testing the replacement. Two apps watching the same key make diagnosis ambiguous. Keep the installed app and saved data until the acceptance checks pass.

Configure the trigger without fighting macOS

Ducduc supports tap-to-start/tap-to-stop and hold-to-record. Check the current key in Settings → Dictation rather than assuming it is fn.

On a fresh setup, fn is the default candidate. If macOS already assigns it to an action such as emoji, input switching or Dictation, Ducduc can choose Right ⌘ Command instead. Other apps and remappers may still use that key; test your actual keyboard.

To use fn, review System Settings → Keyboard → Press fn/🌐 key to and choose Do Nothing if appropriate for your workflow. Ducduc also offers right-hand modifier keys and a recorded custom shortcut. It checks conflicts with its own configured actions, but cannot know every external app’s shortcuts. See who owns the fn key.

Grant permissions, then verify recognition

Ducduc’s setup requests Microphone, Speech Recognition, and Accessibility under System Settings → Privacy & Security. Microphone and speech permissions support recognition; Accessibility supports global triggering and insertion into other apps.

On macOS 14–15, install the required dictation language in Keyboard → Dictation → Languages. The classic Ducduc recognizer requires on-device processing; missing assets do not silently enable server recognition. On macOS 26, its newer path can request speech assets. Apple’s Dictation setup guide covers the system controls.

Start in a simple native text field. If the trigger never starts listening, check the key and Accessibility. If listening starts but no transcript appears, check the microphone, speech permission and language assets. If text exists on a copy card but does not paste, investigate insertion before changing recognition models. Language-asset troubleshooting gives the detailed sequence.

Move vocabulary by meaning, not by file extension

Open Dictionary from the menu bar, or Settings → Cleanup → Manage Personal Dictionary…. Add a term in New word and press Return. Start with a short set of names and technical terms you use daily.

Keep three different mechanisms separate:

  • Vocabulary supplies recognition/cleanup hints. It does not retrain Apple’s recognizer or guarantee every occurrence.
  • Exact replacements are deliberate local rules for whole words or phrases, scoped to All apps or a chosen app.
  • Learned corrections come from eligible edits and spoken corrections. Reading the field, sound/context checks and app scope affect whether learning applies; editing text does not guarantee a saved rule.

A replacement from the old app may be too broad in the new workflow. For example, a hypothetical stage → staging rule might help project notes but damage “stage two.” Scope or omit it rather than importing every mapping blindly. Review accidental learned pairs and remove mistakes. The dictionary guide explains the safeguards.

Once the list is useful, Backup → Export… saves Ducduc terms and correction records. Import merges that backup with existing entries. It is not a complete backup of History, model profiles, API keys and all settings, and it does not create automatic account sync.

Choose the cleanup route before evaluating quality

In Settings → Models, choose built-in cloud, your cloud provider, or local Ollama deliberately.

For local cleanup, verify a downloaded local model and a loopback endpoint, then click Use locally and confirm Offline. A Hybrid backup alone leaves cloud active. Ollama can itself use cloud models, so endpoint naming is not enough. Follow the local setup guide, including disabling Ollama cloud features and testing without network access.

When comparing two apps, use the same microphone, intended language and roughly equivalent cleanup strength. Record raw recognition and final text separately. Otherwise a transcription error and a cleanup change can look like the same failure.

Run a small acceptance test in your real apps

These are sample inputs, not reported test outputs or a claim that every model passes:

CaseWhat to tryAcceptance criterion
Self-correction“Review on Thursday, no, Friday. Do not publish.”Friday and the negative survive
Numbers“The limit is 15, not 50.”No number or polarity change
Mixed languages“我用 Claude Code 写了一个 component。”English terms and Chinese meaning remain
Your namesA short sentence containing saved termsSpellings are useful without invented extra words
DestinationNotes, your browser editor, chat and agent inputCorrect target, formatting and no unintended submission

Then test the recovery paths with harmless text:

  1. Clipboard: copy a marker, dictate with automatic insertion, then paste elsewhere. The prior clipboard should be restored unless you copied something new during insertion. A copy-card Copy is intentional and leaves the dictation on the clipboard.
  2. Changed app: switch apps during dictation. Ducduc should offer the text for copying rather than paste into the unexpected app.
  3. Secure input: test the terminal you use. Secure Keyboard Entry can prevent key monitoring or automatic paste. If recording starts but insertion is blocked, use the copy card. Apple’s Terminal guide explains the security feature.
  4. Offline: after all assets are installed, disconnect every active network path. Confirm whether local cleanup succeeds, or recognized text is preserved with a notice. This tests that run, not all future connected behavior.
  5. Unavailable cleanup: with Offline selected, stop Ollama and verify that text remains recoverable; restart it and retest.

Record the app build, model, settings and failures. Repeat a failed case after changing one setting at a time. A small personal protocol supports your decision; it is not a general accuracy ranking.

Complete the switch only after the important cases pass

If the new app meets your requirements, restore its automatic startup and use the old provider’s billing channel to check renewal or cancellation confirmation. Treat subscription management and app removal as separate tasks. Keep your exports until you know the new dictionary and workflow are complete.

Ducduc free use requires no account; Pro activation uses a license key. The free built-in allowance is 8,000 cleaned words per week, with a 14-day trial beginning at the first built-in cloud cleanup. Current terms and costs are covered in the cost guide.

No account does not mean no cloud processing. History and dictionary remain on the Mac, while cloud cleanup sends relevant text to the chosen service; the receiving app can also sync or submit it. Check the full data route before moving sensitive work.

Try it on your own Mac

Try your own vocabulary and editor, then compare the raw transcript and cleaned text in History.

Download for Mac

Published by the maker of Ducduc. Third-party information follows the sources and dates stated in this guide. Authorship, sources and examples

All posts