Apple Dictation is fine. Here is what a dedicated app adds.
Both free. Both on-device. But EnviousWispr adds AI polish, custom vocabulary, and no timeout. A dedicated dictation app for people who dictate seriously.
Feature comparison
Apple Dictation is a solid default. Here is where a dedicated dictation app pulls ahead.
| EnviousWispr | Apple Dictation | |
|---|---|---|
| Price | Free. No subscription. | Free, built into macOS |
| Setup required | One-time download + model install (~2 min) | None. Already installed. |
| Account required | No | Apple ID (for macOS itself) |
| Audio processing | On-device (Parakeet TDT + WhisperKit) | On-device (Apple Silicon); cloud fallback on Intel |
| Timeout | No inactivity timeout. Handles long dictation sessions, up to a 60-minute cap per recording. | Stops after ~30s of silence |
| AI polish | EG-1, Ollama (on-device), Apple Intelligence (on-device, macOS 26+), or BYO OpenAI/Gemini/Claude key | None |
| Filler word removal | Yes, fixed local list runs even with AI polish off; AI polish removes more | No (transcribes "um", "uh" faithfully) |
| Writing style control | Yes (voice-preserving AI polish that adapts to what you said) | No |
| Custom vocabulary | Custom words with 6-pass fuzzy matching + AI alias suggestions | Not available |
| Transcription speed | 0.61s median; ~1.65s with on-device AI polish | Real-time streaming (words appear as you speak) |
| Accuracy (clean audio) | High (Parakeet TDT 0.6B + AI polish correction) | 90-92% (degrades significantly with background noise) |
| First-word capture | Yes, while the microphone is warm (500ms pre-roll buffer) | Can miss the first word while the mic activates |
| Punctuation | Auto-punctuation + AI polish reformats; spoken punctuation available as an opt-in | Auto-punctuation, emoji dictation, voice commands |
| Voice commands | Spoken emoji (on by default) and spoken punctuation (opt-in); no app navigation or editing commands | "New paragraph", "comma", "smiley face" |
| Clipboard preservation | Yes, when enabled and no other app changes the clipboard first | No (inserts directly at cursor) |
| Auto-stop on silence | Yes (configurable VAD sensitivity) | Stops listening entirely after ~30s silence |
| Transcript history | Yes (searchable transcript log) | No history |
| Language support | 25 European (Parakeet), 99+ languages via WhisperKit | Dozens of locales |
| Accessibility | Keyboard-driven keybind, menu bar app, VoiceOver compatible | Deep OS integration, mixed keyboard + dictation input |
| Works offline | Yes (after one-time model download) | Yes on Apple Silicon; No on Intel |
| Platform | macOS (Apple Silicon required) | macOS, iOS, iPadOS, visionOS |
| Source code | Open source on GitHub (GPLv3) | Closed source |
EnviousWispr latency from production PostHog data (environment=production) on Apple Silicon Macs. Apple Dictation accuracy from published third-party benchmarks. Apple Dictation features verified against Apple Support documentation and firsthand testing on macOS 15. Last verified: August 2026.
What a dedicated dictation app gives you
Apple Dictation works. EnviousWispr works harder. Here is what changes when you move beyond the built-in. For more on why on-device matters, read on-device vs cloud dictation privacy.
Apple Dictation gives you raw transcription. EnviousWispr pipes text through AI polish to remove filler words, fix grammar, and format paragraphs. Choose EG-1, Ollama, Apple Intelligence (macOS 26+), or your own OpenAI, Gemini, or Claude key.
Names, product terms, technical jargon. Apple Dictation has no custom vocabulary feature. EnviousWispr lets you add custom words so "Parakeet TDT" and "Kubernetes" come out right every time.
Apple Dictation stops listening after roughly 30 seconds of inactivity. EnviousWispr has no timeout. Dictate a full paragraph, pause to think, keep going. Sessions can run up to 60 minutes without cutting you off.
Median time from end of speech to transcribed text: 0.61 seconds. Processed locally on the Neural Engine, no network dependency. AI polish, when enabled, adds variable time on top that scales with how much you dictated.
Uh, um, you know, like. Apple Dictation transcribes them faithfully. EnviousWispr's AI polish strips them out so your text reads like writing, not a transcript.
Every line is on GitHub under GPLv3. See exactly how your audio is processed. Apple Dictation is a black box built into the OS.
Dedicated models, dedicated speed
EnviousWispr uses purpose-built ASR models optimized for Apple Silicon. The result is fast, accurate transcription with optional AI refinement.
Based on production data from Apple Silicon Macs. Results vary by hardware and settings.
Where a dedicated app goes deeper
Apple Dictation gives you raw speech-to-text. EnviousWispr adds layers that turn dictation into polished writing.
Apple Dictation has no custom vocabulary feature. If you work with specialized terminology, product names, or technical jargon, you get whatever the built-in model guesses. "Parakeet TDT" might come out as "parakeet teddy." "Kubernetes" might become "Cooper Netties."
EnviousWispr lets you add custom words with a 6-pass fuzzy matching system. Each word can have aliases (phonetic variants the ASR model might produce), and the app suggests aliases automatically using AI analysis of likely misrecognitions. When the model outputs "envious whisper," the custom words system corrects it to "EnviousWispr" before the text reaches your clipboard.
This works at the post-processing layer, so it applies regardless of which ASR backend you use. Add a word once, and it corrects everywhere.
Apple Dictation gives you exactly what you said, including every "um," "uh," "you know," and false start. For quick one-liners, that is fine. For anything longer, you end up editing the transcript to make it readable.
EnviousWispr pipes transcription through an AI polish step that removes filler words, fixes grammar, formats paragraphs, and adjusts tone. You can choose EG-1 (our own on-device model), Ollama (local open-source models), Apple Intelligence (on-device, requires macOS 26+), or bring your own OpenAI, Gemini, or Claude API key.
The polish layer includes hallucination safeguards: short transcripts bypass the LLM entirely, output length is validated against input length, and preamble artifacts ("Certainly! Here is...") are stripped automatically. Your dictated text is wrapped in XML tags with explicit instructions to polish, not answer or rewrite.
Choose Apple Dictation if you only need occasional built-in dictation
Apple Dictation is built into the OS for a reason. It does some things that a third-party app cannot match:
Apple Dictation is already on your Mac. Turn it on in System Settings, press the microphone key, and start talking. No download, no model installation, no permissions to grant. It just works.
iPhone, iPad, Mac, Vision Pro. Apple Dictation follows you across the ecosystem with consistent behavior. EnviousWispr is macOS only and requires Apple Silicon.
Say "new paragraph," "comma," "exclamation mark," or even "smiley face" and Apple Dictation inserts the right character. EnviousWispr supports spoken punctuation (off by default) and spoken emoji, on by default, said as the name followed by the word "emoji" (say "smiley face emoji"), but has no voice commands for app navigation or editing.
Apple Dictation is part of the operating system. No menu bar icon, no extra process, no Accessibility permissions to approve. For light dictation use, the built-in option carries zero overhead.
On macOS 14+, you can type on the keyboard while Dictation is active. Switch freely between voice and keys mid-sentence. EnviousWispr uses a record-then-paste model, so you speak first, then the text appears.
If you dictate more than a few sentences at a time and want polished output with custom vocabulary, give EnviousWispr a try. It runs alongside Apple Dictation with no conflicts.
Common questions
EnviousWispr is a free, on-device alternative that adds AI polish, custom vocabulary, and no timeout limit. It runs alongside Apple Dictation without interfering. Both use Apple Silicon for local processing.
Apple Dictation does not support custom vocabulary. EnviousWispr lets you add custom words for names, brands, and technical terms. The app applies them during transcription so specialized words come out correctly.
Apple Dictation has a built-in inactivity timeout. If you pause speaking for too long, it stops. EnviousWispr has no inactivity timeout, so you can pause to think and continue dictating without restarting. Each recording does have a generous one-hour cap.
It runs alongside Apple Dictation, not instead of it. You can keep Apple Dictation enabled for quick one-liners and use EnviousWispr for longer dictation where you want AI polish and custom words. They use different keybinds and do not conflict.
Yes. No subscription, no usage limits, no account required. Download and use it. The source code is available on GitHub under a GPLv3 license.
Transcription runs entirely on-device after a one-time model download. AI polish can also run locally via EG-1, Ollama, or Apple Intelligence (macOS 26+). No internet required for the full pipeline.
EnviousWispr uses Parakeet TDT, a modern ASR model, combined with AI polish that fixes misrecognitions and grammar. Apple Dictation achieves ~90% in quiet conditions (third-party tests) in clean environments but degrades significantly with background noise. EnviousWispr's AI polish layer provides an additional correction pass that Apple Dictation lacks.
Yes. EnviousWispr removes spoken noises like "um," "uh," and "hmm" automatically, on a fixed local list that runs even with AI polish switched off. AI polish defaults to Apple Intelligence, which needs macOS 26+; below that it silently falls back to the same deterministic cleanup unless you switch to EG-1, Ollama, or a cloud key. When AI polish does run, it goes further and also removes false starts and repeated words. Apple Dictation transcribes them as-is. If you dictate longer passages, filler removal makes a noticeable difference in readability.
Not the same way. Apple Dictation supports voice commands such as "new paragraph," "comma," and emoji names as its main punctuation method. EnviousWispr supports spoken emoji (on by default) and spoken punctuation (opt-in, since AI polish already handles most punctuation automatically), but has no voice commands for app navigation or editing.
Any Mac with Apple Silicon (M1 or later) running macOS 14 Sonoma or newer. The Neural Engine on Apple Silicon is what makes on-device transcription fast. See the 2-minute getting started guide.
Yes. They use separate keybinds and separate audio pipelines. Keep Apple Dictation for quick inline corrections and voice commands. Use EnviousWispr for longer dictation where you want AI polish, custom vocabulary, and transcript history. They do not interfere with each other.
Try free dictation that works in any Mac app
Free to download. No account required. Runs alongside Apple Dictation.