EnviousWispr vs VoiceInk: free on-device Mac dictation
Both keep your audio on your Mac. EnviousWispr adds a faster speech engine, a deeper AI polish pipeline, and costs nothing.
The short answer
Is VoiceInk free, and is it open source?
VoiceInk is a Mac dictation app that can transcribe on your Mac, depending on which of its three engines you pick. It is open source under GPL v3, and the app is paid: one-time pricing of $29 for one Mac, $49 for two, or $69 for three, with no subscription.
- What it costs
- $29 Solo, $49 Personal, $69 Extended. One-time, not a subscription.
- Is it free?
- The source is public, so you can build it yourself. The packaged app is paid.
- Is it open source?
- Yes, GPL v3 on GitHub. Same licence EnviousWispr uses.
- Where audio goes
- On device with whisper.cpp or Parakeet. Its third engine, Cohere Transcribe, is a cloud option, so that one sends audio out.
EnviousWispr is free, GPL v3, and always transcribes on your Mac, with no engine that sends audio anywhere. The full comparison is below. Pricing and engines re-verified from tryvoiceink.com and the VoiceInk GitHub repository on 21 August 2026.
Feature comparison
An honest look at how the two on-device dictation apps compare across pricing, speed, engines, and features.
| EnviousWispr | VoiceInk | |
|---|---|---|
| Price | Free. No payment ever. | Solo $29 / Personal $49 / Extended $69, one-time* |
| Account required | No | No |
| Audio processing | On-device (Apple Silicon) | On-device with whisper.cpp or Parakeet; Cohere Transcribe is a cloud option |
| Audio leaves your Mac | Never. If you enable cloud AI polish, only the text transcript is sent. | Not with whisper.cpp or Parakeet. Selecting Cohere Transcribe likely sends audio to Cohere's API. |
| Works offline | Yes (after model download) | Yes (after model download) |
| Speech engines | Parakeet TDT + WhisperKit | whisper.cpp, Parakeet (FluidAudio), or Cohere Transcribe, user-selectable |
| AI polish | 6 providers: EG-1, Ollama (on-device), Apple Intelligence (on-device, macOS 26+), OpenAI, Gemini, Claude (BYO key) | AI formatting via system prompts with LLM providers |
| Offline AI polish | Yes (EG-1, Ollama, or Apple Intelligence on macOS 26+, no API key needed) | Requires cloud LLM for AI formatting |
| Filler word removal | Yes (built-in regex pass, runs before AI polish) | Via AI formatting prompts |
| Writing style control | Voice-preserving AI polish that adapts to what you said | Power Mode with per-app custom prompts |
| Custom vocabulary | 6-pass fuzzy matching (Levenshtein + Soundex + bigram) | Personal dictionary (find-and-replace) |
| AI hallucination safeguards | Yes (length validation, preamble stripping, short-text bypass) | Not documented |
| First-word capture | Pre-roll buffer captures audio before you finish pressing the keybind | Standard recording start |
| Starts listening instantly | Warm engine (model stays loaded between recordings) | Model loaded at startup |
| Clipboard preservation | Yes, when enabled and no other app changes the clipboard first | Not documented |
| Text lands in the right app | Yes (three-tier paste with target app reactivation) | Pastes into active app |
| Auto-stop on silence | Yes (VAD-based silence detection) | Yes |
| Multi-language | 25 European (Parakeet), 99+ languages via WhisperKit | 100+ languages |
| Per-app settings | Global keybind and settings | Power Mode with per-app prompts |
| Screen context | No | Yes |
| Accessibility | VoiceOver labels, keyboard-navigable settings | Standard macOS controls |
| Transcription latency | 0.61s median; ~1.65s with on-device AI polish | Depends on model size and hardware |
| Source code | Open source (GPLv3) | Open source (GPL v3) |
| Platform | macOS (Apple Silicon, 14+) | macOS (14+) |
*VoiceInk pricing restructured 2026-08-01: Solo $29 (1 Mac), Personal $49 (2 Macs, most popular), Extended $69 (3 Macs), all one-time/lifetime, no subscription. Re-verified live against tryvoiceink.com. EnviousWispr latency from production PostHog data on Apple Silicon Macs. VoiceInk claims source-verified from tryvoiceink.com and GitHub; not firsthand-tested. Competitor claims last verified: 2026-08-21.
Why choose EnviousWispr over VoiceInk
Both apps keep your audio private. Here is where EnviousWispr pulls ahead.
No tiers, no license keys, no payment. VoiceInk charges a one-time fee starting at $29 (Solo tier, 1 Mac). EnviousWispr costs nothing.
Parakeet TDT covers 25 European languages and runs on Apple's Neural Engine. 0.61s median latency. WhisperKit handles 99+ languages when you need wider coverage.
EG-1 and Ollama run fully on-device with no API key, and so does Apple Intelligence on macOS 26+. Or bring your own OpenAI, Gemini, or Claude key. VoiceInk offers AI formatting through system prompts with cloud LLMs.
Parakeet TDT for speed on English by default. Switch to WhisperKit in Settings for multilingual coverage. See how the pipeline works. VoiceInk offers whisper.cpp, Parakeet, and Cohere Transcribe, also a manual pick between engines in its own settings.
A 6-pass fuzzy matching system catches misrecognized words that simple find-and-replace would miss. Levenshtein distance, Soundex phonetic matching, and bigram similarity work together to correct names, jargon, and technical terms.
Every line is on GitHub under GPLv3. Verify privacy claims yourself. VoiceInk is also open source under GPL v3, so both apps offer transparency.
Where does your voice go?
Both apps process audio on your Mac. The difference is in how polish and formatting work. For a deeper dive, read on-device vs cloud dictation privacy.
Tuned default vs pick-your-own
Both apps can run Parakeet on-device now, but the setup differs. EnviousWispr defaults to Parakeet TDT, tuned for Apple's Neural Engine, so you get the speed without configuring anything. VoiceInk offers Parakeet, whisper.cpp, and Cohere Transcribe as options you choose yourself.
EnviousWispr latency based on production data from Apple Silicon Macs. VoiceInk latency depends on chosen Whisper model size and hardware. Results vary by settings.
The details that make dictation reliable
On-device transcription is only half the problem. What happens between speech and pasted text determines whether dictation feels trustworthy.
Dictation apps face a subtle problem: between when you stop recording and when text is pasted, you might switch apps, click a different field, or have something important on your clipboard. If the app does not account for this, your text ends up in the wrong place or overwrites what you had copied.
EnviousWispr captures which app had focus when you started recording. While the microphone is warm, a 500ms pre-roll audio buffer reduces first-word clipping; on a cold start, there is no earlier audio to draw on and the first word or two can still be lost.
After transcription, the app re-activates the original target app and tries direct text insertion first, then an ordinary paste, then the app's own Edit > Paste menu. If all three fail, your text stays on the clipboard so you can paste it yourself. When clipboard restoration is on (the default), your prior clipboard contents go back after a clipboard-based paste, unless another app changed the clipboard in between.
Every speech-to-text engine struggles with names, acronyms, and domain-specific terms. Simple find-and-replace dictionaries only work when the ASR output exactly matches the expected error. But ASR errors are phonetic: "Envious Whisper" instead of "EnviousWispr," or "park it" instead of "Parakeet."
EnviousWispr runs custom vocabulary through a 6-pass fuzzy matching pipeline. First, exact matches are applied. Then Levenshtein distance catches close misspellings. Soundex phonetic encoding finds words that sound alike but are spelled differently. Bigram similarity catches partial matches. Case-insensitive and word-boundary passes handle the remaining edge cases.
The result: you add "EnviousWispr" to your custom words once, and it corrects "envious whisper," "envious wisper," and "envious whispr" automatically. VoiceInk's personal dictionary is useful for exact replacements, but phonetic misrecognitions require the fuzzy matching approach.
Choose VoiceInk if these matter more to you
VoiceInk is a solid, lightweight dictation app. It may be a better fit in these situations:
VoiceInk is a one-time purchase, from $29 for one Mac up to $69 for three. You pay once and own it. EnviousWispr is free today, but as a younger project its long-term model is still evolving. If you prefer the certainty of a single purchase, VoiceInk is straightforward.
VoiceInk's Power Mode lets you configure different prompts and behaviors for each application, and can read what is on your screen to provide more relevant formatting. EnviousWispr uses global settings and does not have screen context awareness.
VoiceInk does one thing well with a clean interface and minimal setup. EnviousWispr has more moving parts (dual engines, multi-provider AI polish, fuzzy vocabulary). If you want the simplest possible dictation tool, VoiceInk may feel more focused.
VoiceInk is licensed under GPL v3, an OSI-approved open source license. EnviousWispr is also licensed under GPL v3, so both apps are open source under the same license.
VoiceInk supports 100+ languages through whisper.cpp and has been tuned for multilingual workflows. EnviousWispr supports 99+ via WhisperKit, and Parakeet TDT (the fast engine) covers 25 European languages. If you dictate heavily in non-European languages, VoiceInk may be more polished for that use case.
VoiceInk has a straightforward download-and-go experience. EnviousWispr requires granting Accessibility permissions and downloading an ASR model on first launch. Both are fast, but VoiceInk may feel quicker to get started with.
If you primarily dictate in English and want the fastest transcription at zero cost, give EnviousWispr a try. If VoiceInk's per-app settings or screen context awareness are important to your workflow, it is worth the price.
Common questions
Yes. EnviousWispr has no purchase price, no subscription, and no usage limits. VoiceInk charges a one-time fee starting at $29 (Solo tier).
Different engines have different speed and accuracy profiles. EnviousWispr uses Parakeet TDT by default, a CTC-Transducer model optimized for Apple's Neural Engine, which delivers 0.61s median latency for English. VoiceInk also now offers Parakeet (via FluidAudio) and Cohere Transcribe alongside whisper.cpp as selectable engines, so this is no longer a Parakeet-vs-whisper.cpp distinction; EnviousWispr defaults to Parakeet and lets you switch to WhisperKit in Settings when you need broader language coverage, the same manual-switch shape as VoiceInk's own engine picker.
Yes. Parakeet TDT covers 25 European languages. For wider coverage, EnviousWispr switches to WhisperKit, which supports 99+ languages. VoiceInk supports 100+ languages through whisper.cpp.
EnviousWispr uses a 6-pass fuzzy matching system with Levenshtein distance, Soundex phonetic matching, and bigram similarity to catch misrecognized words even when the ASR output is phonetically close but spelled differently. VoiceInk offers a personal dictionary with find-and-replace rules, which works well for exact substitutions but may miss phonetic errors.
AI polish is a post-transcription step that fixes grammar, removes filler words, and formats your text using a language model. EnviousWispr supports EG-1, Ollama, and Apple Intelligence (all on-device, Apple Intelligence needs macOS 26+), and cloud providers (OpenAI, Gemini, Claude) with your own API key. VoiceInk offers AI formatting through system prompts with cloud LLM providers.
VoiceInk supports 100+ languages through whisper.cpp and has been optimized for multilingual use. EnviousWispr supports 99+ via WhisperKit, and its fastest engine (Parakeet TDT) covers 25 European languages. For languages outside those, VoiceInk may have an edge in breadth and tuning.
Yes. VoiceInk offers AI formatting through system prompts with LLM providers. EnviousWispr supports 6 AI polish engines including EG-1 (our own on-device model), Ollama (on-device), Apple Intelligence (on-device, macOS 26+), OpenAI, Gemini, and Claude, plus built-in hallucination safeguards that reject fabricated output.
Yes. Download EnviousWispr, set your keybind, and start dictating. There is no data to migrate. Both apps work in any text field on macOS. See the 2-minute getting started guide.
EnviousWispr requires Apple Silicon (M1 or later) running macOS 14 Sonoma or newer. VoiceInk also requires macOS 14+. Both apps target modern Macs.
Yes. EnviousWispr is a free, on-device dictation app for Apple Silicon Macs. It uses Parakeet TDT for fast English transcription and includes an AI polish pipeline with 6 providers. No account or payment required.
Open source under the GNU General Public License v3 (GPLv3), the same OSI-approved license VoiceInk uses. You can read, build, and inspect every line of code on GitHub.
Ready to try free on-device dictation?
Free to download. No account required. Faster than whisper.cpp for English.