Choosing a Mac Voice Engine: Parakeet, Whisper, Cloud, and Apple

Seven choices, not two

The Parakeet versus Whisper framing is too small for what the Mac app actually offers. Auri Voice gives you seven Voice Engine choices: five local models you download once, and two that need no download at all. Every one of them is a setting you can change in about ten seconds, which is the most useful fact in this guide.

The picker shows a relative speed and accuracy indicator beside each local engine. Read those as rankings against each other rather than scores you can compare to a published benchmark. Labs may also expose experimental engines such as Parakeet Live and Cohere Transcribe; treat anything in Labs as early and changeable.

Question one: how many languages do you speak into it

Language coverage decides more engine choices than accuracy does, and it is the one question you can answer without any testing. The two Parakeet engines exist precisely because that trade is real.

• Only English, ever: Fast English Dictation (Parakeet TDT v2). It rates higher on accuracy than the multilingual option and gives up every other language to get there. This is the Auri Voice pick for English speakers, and most people should start here.
• Two or more languages in a normal week: Fast Multilingual Dictation (Parakeet TDT v3). Slightly lower on the accuracy indicator, and it stops you from switching engines every time you answer a message in another language.
• A long tail of languages: Auri Cloud covers 100+ and needs no download, which matters when the language you need appears twice a year.

Be honest about the second bullet. People who dictate one foreign message a month often do better on the English engine plus a manual fix, while people who genuinely live in two languages lose more time to engine switching than they ever gain from the accuracy difference. Try v2 for a week first; if you find yourself in the settings panel, move to v3 and stop thinking about it.

Question two: what your Mac can carry

Local models cost disk, memory, and warmth. Those costs are invisible in a comparison table and very visible on a fanless laptop at 40 percent battery in the afternoon.

• Whisper Large Turbo is the high-quality local option. Reach for it when audio is genuinely difficult: a strong accent, second-language speech, a room with constant noise, or vocabulary that general models rarely see. It wants headroom, so it suits a desk Mac better than a thin one on battery.
• Whisper Base is the opposite end. It runs on almost any Mac and asks for very little disk. Choose it for an older machine, a nearly full SSD, or a second Mac you use occasionally.
• Nemotron 3.5 ASR targets the gap between speaking and seeing text. If latency is the thing that breaks your writing rhythm rather than word errors, try it before you try a bigger model.

A split setup is common and costs nothing to maintain: Whisper Large Turbo for drafting sessions at the desk, a Parakeet engine for replies on battery. Since the engine is one setting, you can change it as often as you change the room you work in.

When no download is the right answer

Two options skip the download entirely, and each solves a different problem.

Auri Cloud is fast, accurate, and covers 100+ languages without putting a model on your SSD. It fits a Mac with little free space, a machine you use rarely, or work that spans languages you cannot predict. The trade is plain: audio leaves the device, so it is the wrong choice for material your employer or client would not want off the machine.

Apple On-Device is built into macOS and costs nothing to enable. Use it to confirm your microphone, permissions, and insertion path all work before you commit disk to anything else. It is a reasonable floor rather than a destination, since a downloaded engine plus custom vocabulary handles long takes and proper nouns more comfortably.

• Little free disk, many languages: Auri Cloud.
• Sensitive material, no network: a downloaded local engine.
• Just checking the plumbing works: Apple On-Device.
• Flights and locked-down offices: download before you leave.

The engine is not the whole pipeline

Choosing an engine settles how audio becomes words. It does not settle what happens to those words next. Auri Voice runs a separate text improvement pass that removes filler words, resolves false starts so your final intent survives, adds punctuation, and applies formatting. That pass has its own setting, called AI Processing.

• Automatic (recommended): on-device AI when your Mac supports it, Auri Cloud otherwise.
• Auri Cloud: always the hosted path.
• Bring your own key: OpenAI, Anthropic, xAI, Groq, Cerebras, Google, OpenRouter, Apple Intelligence where available, or a custom OpenAI-compatible endpoint you run yourself.

Read the two settings together rather than separately. A local engine paired with Automatic on a Mac without on-device AI support means your text goes out for cleanup even though your audio never did. If nothing may leave the machine, pin both sides. If you are already on a cloud engine, the AI Processing choice is about model quality and cost, not about privacy.

This also changes which errors should influence your engine choice. Fillers, restarts, and missing commas get fixed for you. Wrong names, wrong numbers, and homophones that read perfectly do not. Judge engines on the second group.

A bake-off with an email you already sent

Published error rates come from read-aloud recordings in quiet rooms. Your accent, your microphone, your office, and the twelve proper nouns you say every day are all missing from them. Twenty minutes with your own material beats any table, including the one at the top of this guide.

1. Open a long email you actually sent last week, ideally one with two names, a date, and a number in it.
2. Read it aloud into a draft once per candidate engine, same microphone, same room, same time of day.
3. Compare each result against the original and count two things only: words you would have to fix, and seconds you waited after you stopped talking.
4. Add your recurring names and product terms to custom vocabulary, which works with the Parakeet engines, then rerun the leader.
5. Keep the winner for a full week of real work before you open settings again.

Before you conclude anything, fix the inputs. Move the microphone to six or twelve inches instead of a laptop lid two feet away, stop switching between AirPods and the built-in mic, silence the desk fan, and speak full sentences at your normal pace. Most people find the correction count drops far enough that the engine question stops being interesting, which is the outcome you want. Save the comparison for the errors that survive a good setup.

FAQ