K

    KB-Whisper

    A collection of Swedish-optimized speech-to-text models based on the Whisper architecture, developed by the National Library of Sweden's KBLab.

    SESweden
    Headquarters
    πŸ‡ΈπŸ‡ͺSweden
    Data Hosting
    United States
    Structure
    Independent
    Compliance
    β–£GDPRβ—†EU HQ

    KB-Whisper key facts

    Last updated
    Legal entity
    Kungliga biblioteket (National Library of Sweden)
    Headquarters
    Sweden (SE)
    Ownership
    Part of National Library of Sweden (Kungliga biblioteket)
    Data hosting
    United States
    Infrastructure
    Hugging Face
    Open source
    No
    US Cloud Act exposure
    Exposed
    Replaces
    OpenAI Whisper, Google Cloud Speech-to-Text, Azure Speech Services, AssemblyAI

    About KB-Whisper

    KB-Whisper is a collection of specialized speech-to-text models hosted on the Hugging Face platform, developed by KBLab, the data lab of the National Library of Sweden. These models leverage OpenAI's Whisper architecture but are specifically fine-tuned using the library's extensive Swedish language archives to provide superior performance for Swedish audio transcription. As an open-source initiative from a Swedish state authority, the project aims to facilitate digital sovereignty by providing high-quality, transparent AI tools for the Swedish language. The models are publicly accessible and intended for researchers, developers, and public sector organizations who require robust transcription capabilities without relying solely on proprietary US-based commercial APIs.

    Alternative to

    Hosting & Security

    Infrastructure

    Hugging Face

    Key Features

    Pre-trained speech recognition models
    Fine-tuned Whisper architecture
    Open-source model weights
    Swedish language optimization
    Integration with Hugging Face transformers library

    Related Articles

    All articles