Parakeety for writers and researchers
Most professional writers and researchers have tried voice dictation at least once and bounced off it for the same reason: the transcript that comes back is not quite right, and fixing it takes longer than just having typed it. The case for trying again is that the models have got dramatically better in the last two years, and the apps wrapping those models have got faster. Parakeety is one of those apps: local push-to-talk dictation for Mac, no cloud round-trip, no language flag to set. Hold the section key, talk through a paragraph, release; the words appear at the cursor in whichever document you are writing in. The first-draft writing speed is not the same as typing. It is faster than typing, by a comfortable margin, once your brain stops rephrasing for the keyboard.
First-draft workflows
The honest answer about where dictation fits is: not all writing tasks. Editing, revising, and structural work belong to the keyboard and the cursor; you need to see what you are moving around. Dictation earns its keep on first drafts, where the bottleneck is generating words rather than arranging them.
That maps to specific moments in a writer's day:
- Novelists and short-fiction writers. Pacing a scene aloud, voicing dialog, walking through a character's interior monologue. The spoken cadence often matches what you want on the page better than the typed one.
- Journalists. Drafting a quick reaction piece, dictating the framing paragraphs of a longer feature, getting first words into a Google Doc before they evaporate.
- Content writers and bloggers. Outlines, intros, and the painful middle paragraphs that nobody actually wants to type.
- Academics and researchers. Drafting paper sections, dictating notes against a paper you are reading, writing up findings before formalizing them.
In all of these the dictation is the first pass; the editing afterwards is what shapes the final work. Parakeety is built for that first pass. Local dictation for engineers covers the first-draft case for engineering work, in commit messages, design docs and code-review comments.
Researchers and multilingual source quotes
The 25-language list (English, French, German, Spanish, Italian, Portuguese, Dutch, Polish, Czech, Slovak, Hungarian, Romanian, Bulgarian, Croatian, Slovenian, Greek, Swedish, Danish, Finnish, Estonian, Latvian, Lithuanian, Maltese, Russian, Ukrainian) covers most European research literature. The auto-detection is what makes it usable. You can dictate an English paragraph framing a quote, switch to dictating the German source verbatim, switch back to English commentary, all without changing a setting. The model picks up which language you are speaking by itself.
For researchers working with translated or partially translated material this is the bit that earns the price. Whisper-based apps need an explicit language flag and tend to mistranslate or anglicize foreign-language source quotes; Parakeet TDT v3 handles the switch in stride. Languages outside the supported list, including Mandarin, Japanese, Korean, Arabic, Hindi, Turkish and Hebrew, are not currently supported. If your source material is in those languages, this is not the tool yet. Local dictation for translators covers the same multilingual flow from the professional translation-and-interpretation angle.
Why on-device matters for unpublished work
There is a category of writing whose value depends on it not leaking before publication: an unfinished novel, an embargoed feature, a research paper before peer review, a memoir naming names. Cloud dictation makes that category complicated. Most reputable services say they do not store the audio long-term and do not train on it, but the audio does pass through their servers, and transcripts are usually retained for some period to support the product. Whether that is compatible with your contract, your publisher's policy, or your own discretion is your call.
Parakeety side-steps the question. Audio is captured in memory while the key is held, processed on the Apple Neural Engine, the transcript pastes at the cursor, the audio buffer is discarded. Nothing is written to disk, nothing leaves the Mac. For writers under embargo or with non-disclosure obligations against an editor or institution, the architectural answer is cleaner than the contractual one.
What Parakeety is not
Worth being honest. Parakeety is push-to-talk: hold the key, talk, release. It is not designed for transcribing recorded interviews, lectures, or hour-long files. There are good Whisper-based tools for that workflow (MacWhisper is the obvious one). Parakeety does not try to compete with them. It also does not do AI cleanup or punctuation rewriting; the transcript that pastes is the transcript the model produced, with the punctuation it inferred. That tends to be good for sentence-level prose and middling for paragraph-spanning thought; you will probably end up adding your own paragraph breaks in the edit pass. For the first-draft job it is built for, that trade-off is fair.
FAQ
- Does it punctuate properly for long-form prose?
- Mostly yes for sentences and reasonably yes for paragraphs. The model infers punctuation from intonation and pauses, so it tends to put commas and full stops in roughly the right places. Where it tends to be lighter is on paragraph breaks across longer passages. For those you will typically add the breaks during the edit pass. Em-dashes and semicolons it does not produce; those you also add by hand. For most working writers that is a fair trade for first-draft speed.
- Can I use it to transcribe recorded interviews or long audio files?
- No. Parakeety is push-to-talk only. You hold a key, dictate live, and release; it does not accept audio files as input. If your need is to transcribe an existing recording, MacWhisper is the conventional Mac tool for that workflow. The two solve different problems.
- Can I dictate research source quotes in another European language?
- Yes, across the 25 supported languages, with auto-detection. You can dictate an English paragraph, switch to dictating a German or French source verbatim, switch back to English, without changing any setting. The model identifies the language by itself. Languages outside the supported list, including Mandarin, Japanese, Korean, Arabic, Hindi, Turkish and Hebrew, are not supported at the moment.
- Is it suitable for embargoed or unpublished work?
- Architecturally yes, contractually it depends on your specific obligations. Audio is processed on the Mac and discarded after transcription; nothing is written to disk and nothing is uploaded. The only outbound calls Parakeety makes are the one-time speech-model download on first launch and periodic license checks (which carry only the license key, a hardware ID hash, and the machine hostname). For embargoes that simply require the audio not to leave your machine, that is a clean answer. For embargoes with specific tooling clauses, check those clauses against the architecture.
Try it
Parakeety is a Mac menu-bar app. Hold the section key, talk, release; your words paste at the cursor in whichever app you were typing into. Audio never leaves the machine. There is a free 7-day trial with no card required. After that it is $30 once.