🇸🇪 Läs på svenska →

Research · Swedish Dictation

KB-Whisper for Swedish Dictation on Mac: Why It Matters

EN Updated Fredrik Carlsson 5 min read

OpenAI’s Whisper is a remarkable achievement. Trained on 680,000 hours of audio from across the internet, it handles dozens of languages without any language-specific configuration. For Swedish, this means it can transcribe common speech with reasonable accuracy. But “reasonable” for a general-purpose multilingual model is not the same as “built for Swedish.” The National Library of Sweden decided to close that gap.

KB-Whisper is the result: a fine-tuned Whisper model trained specifically on Swedish speech, developed by KBLab, the AI lab at Kungliga biblioteket, the Royal Library of Sweden. This article explains how it was built, what makes it better for Swedish dictation, and how it runs locally on a Mac with no audio leaving your device.

What KB-Whisper is

KB-Whisper is not a different architecture from Whisper: it uses the same transformer-based design. What sets it apart is the training data. KBLab fine-tuned the model on over 50,000 hours of specifically Swedish speech, drawn from:

This speech spans domains, speakers, and registers that matter for professional dictation: formal and informal speech, legal language, political discourse, technical explanations. Generic Whisper was exposed to some of this via internet audio, but Swedish is a small fraction of its 680,000-hour training corpus. KB-Whisper inverts that ratio for Swedish specifically.

47%
Fewer word errors on Swedish
vs whisper-large-v3
50k+
Hours of Swedish
training audio

Both figures come from KBLab, not from me. The 47% is KBLab’s own measurement of KB-Whisper against OpenAI’s whisper-large-v3, averaged over three Swedish test sets (FLEURS, CommonVoice and NST) and published at Interspeech 2025. It applies to Swedish only.

Why generic Whisper struggles with Swedish

Swedish has specific properties that challenge a general-purpose multilingual model:

Compound words. Swedish compresses concepts into single words in ways English rarely does: samhällsutvecklingen (society’s development), arbetsmarknadsutbildning (labor market training), dataskyddsförordningen (the data protection regulation). A model not well-trained on Swedish will frequently split these incorrectly or substitute a near-English equivalent.

Swedish proper nouns. Place names like Östersund, Hässleholm, Ånge, Luleå; names like Björn, Åsa, Gösta. These are phonetically distant from English-language patterns, and generic Whisper approximates them inconsistently.

Professional vocabulary. Legal, medical, technical, and political terms in Swedish differ substantially from their English counterparts. A Swedish court document contains words like rättegångsbalken, förundersökning, säkerhetsställning, none of which map cleanly onto English training data.

NB-Whisper for Norwegian

The Norwegian National Library (Nasjonalbiblioteket) has done equivalent work for Norwegian. NB-Whisper, developed by NbAiLab, is fine-tuned on Norwegian taledata from NRK, Stortinget, and archival sources. Both KB-Whisper and NB-Whisper run locally in Sæga. You can switch between Swedish and Norwegian models in Settings. Read about NB-Whisper and Norwegian dictation →

How KB-Whisper runs locally on Mac

Whisper models need hardware acceleration to answer quickly. Apple Silicon has it built in: the M-series chips have a Neural Engine and GPU cores that models can use through Metal. KB-Whisper runs on the GPU cores, and the text arrives a moment after you stop talking.

In Sæga, the transcription flow is entirely local. When you press the hotkey and speak, the audio is captured, processed on your device by the speech model you picked, and the text is inserted at your cursor. In Raw mode nothing, neither audio nor text, leaves your Mac. If you dictate client notes or patient details, that is often the requirement rather than a preference.

KB-Whisper in practice: what actually improves

The 47% is an average. In practice it shows up in a few places:

Size matters within the family too. When I measured KB-Whisper myself on 1,804 Swedish dialect clips, large got 83.9% of the clips entirely right and small 70.1%. The full measurement, with raw data →

Who builds this and why it is not a product

KBLab is the AI research lab at Kungliga biblioteket, Sweden’s national library. Their mandate is to advance AI for Swedish language and culture, not to ship commercial products. KB-Whisper is published as a research contribution: the weights are available for researchers, developers, and anyone who wants to use them.

This is important context. KB-Whisper does not compete with Whisper: it is built on top of it, with the explicit goal of serving Swedish speakers better. NbAiLab at the National Library of Norway has done the same for Norwegian with NB-Whisper.

How to try KB-Whisper for Swedish dictation

Sæga runs KB-Whisper for Swedish and NB-Whisper for Norwegian, next to Pianissimo, the Swedish default. To try KB-Whisper:

Everything that runs on your Mac is free, with no cap: dictation, every speech model, the dictionary, voice commands and history. Pro is $29 once, no subscription, for what goes through the cloud: 1,000 AI texts a month, custom modes for every app, Email, AI Prompt and Rewrite.

Then dictate something you would write anyway, say a note to your Arbetsförmedlingen case worker about arbetsmarknadsutbildning in Trollhättan, once with Pianissimo and once with KB medium, and keep whichever leaves you less to fix. Every Mac dictation app I looked at (Swedish) →

Try Swedish dictation free, no account needed

Sæga runs Pianissimo as the Swedish default, with KB-Whisper one click away. Both run on your Mac, both are free, and no audio is sent anywhere.

Download Sæga, free to start
macOS 14 Sonoma or later · Apple Silicon

Compare Free and Pro →