W

Wav2vec2 Large Voxpopuli Sv Swedish

Developed by KBLab
This model is based on Facebook's VoxPopuli-sv large model, additionally pre-trained and fine-tuned using Swedish radio programs, NST, and Common Voice data.
Downloads 38.78k
Release Time : 3/2/2022

Model Overview

An automatic speech recognition (ASR) model for Swedish, based on the Wav2vec 2.0 architecture, trained and fine-tuned on various Swedish datasets.

Model Features

Multi-dataset training
Pre-trained and fine-tuned on Swedish local radio programs, NST, and Common Voice datasets
High performance
Achieves a WER of 3.95% on the NST + Common Voice test set and 10.99% on the Common Voice test set
Supports language model integration
Using a 4-gram language model reduces WER on the Common Voice test set from 10.99% to 7.82%

Model Capabilities

Swedish speech recognition
16kHz audio processing

Use Cases

Speech-to-text
Radio program transcription
Automatically transcribe Swedish radio programs into text
WER 3.95% (on NST + Common Voice test set)
General speech recognition
Convert Swedish speech to text
WER 10.99% (on Common Voice test set)
Featured Recommended AI Models
AIbase
Empowering the Future, Your AI Solution Knowledge Base
Š 2025AIbase