What Is Whisper and Why Is It Everywhere?

Short answer: Whisper is OpenAI's open-source speech-to-text model, released in 2022. It is accurate, handles dozens of languages, and is free for developers to run locally. That combination is why it powers so many Mac dictation apps, including on-device tools like BlaBlaType that never upload your voice.

What Whisper actually is

Whisper is an automatic speech recognition model: software that listens to audio and writes down the words. OpenAI released it in late 2022 and, crucially, published the model weights under a permissive open-source license. That means any developer can download it and run it inside their own product without paying per minute. You can read OpenAI's original research write-up in the Whisper paper on arXiv if you want the technical detail.

Two things made it stand out immediately. First, accuracy: it was trained on a very large and varied set of audio, so it copes well with accents, background noise, and casual speech. Second, breadth: it was trained on multilingual audio, so a single model can transcribe many languages and even translate into English. Before Whisper, that level of quality usually meant a paid cloud service.

Why Whisper is suddenly everywhere

The short version is that Whisper removed the two biggest barriers to building good voice software: cost and access. When the best speech engine is free and open, a solo developer can ship a dictation app that rivals what only big companies could offer before. So they did, by the dozen.

• It is free to run. No per-word cloud bill, which changes the entire economics of a dictation app.
• It runs locally. On a modern Mac, Whisper is fast enough to transcribe as you speak, so nothing has to leave the machine.
• It is multilingual. One model covers dozens of languages, which is expensive to build from scratch.
• It is open. A whole ecosystem of optimized versions, like the popular community ports, grew around it.

There is also a distinction worth keeping straight. OpenAI's better-known chat and voice products, documented in its voice mode FAQ, run in the cloud. The Whisper model, by contrast, can run entirely offline on your own device. That difference is the whole reason on-device dictation is even possible.

How Whisper turns your voice into text

You do not need to understand the internals to use it, but the pipeline is simpler than it sounds. Here is what happens between you speaking and clean text appearing.

Whisper, Apple Dictation, and the newer models

Whisper is not the only speech engine, and the differences are practical rather than academic. The table below is a fair, high-level comparison of the common options you will run into on a Mac.

Engine	Open-source	Runs on-device	Multilingual	Typical use
Whisper	Yes	Yes	Dozens	Dictation and transcription apps
Parakeet	Yes	Yes	Growing	Fast real-time dictation
Apple Dictation	No	Mixed	Many	Built into macOS
Cloud ASR APIs	No	No	Many	Server-side transcription

This is why BlaBlaType uses local models like Whisper and Parakeet together: Whisper brings robust, multilingual accuracy, and Parakeet adds speed for real-time typing. Both run on your Mac, so your audio and transcripts never leave the device. If privacy is your main concern, we cover exactly where your voice goes in is Mac dictation private.

The engine is not the whole story

Here is the part most comparisons miss. A great model gives you an accurate raw transcript, but raw speech is messy: it has filler words, false starts, and no real punctuation. What makes voice typing genuinely useful is the layer on top: cleaning that text up and dropping it wherever your cursor is. That is why two apps using the same Whisper model can feel completely different to use.

BlaBlaType pairs local speech recognition with on-device AI cleanup powered by Apple Intelligence, so filler disappears, punctuation and grammar get fixed, and tone can be adapted, all without a server round-trip. A custom dictionary keeps names and jargon correct. Since most people speak around three to four times faster than they type, a clean pipeline turns that raw speed into finished text. Curious what the AI layer does exactly? See how tiered dictation apps package these features, or if you write code, how voice coding actually works on a Mac. Pricing lives on our plans page.