W

Wav2vec2 Base Fi Voxpopuli V2 Finetuned

Developed by Finnish-NLP
A Finnish automatic speech recognition model fine-tuned based on facebook/wav2vec2-base-fi-voxpopuli-v2, trained with 276.7 hours of annotated data, supports KenLM language model decoding
Downloads 64
Release Time : 5/14/2022

Model Overview

Speech-to-text model optimized for Finnish, performs excellently on test sets like Common Voice

Model Features

Efficient fine-tuning
Based on the VoxPopuli V2 pre-trained model, fine-tuned with 276.7 hours of Finnish data
Multi-dataset support
Incorporates 6 data sources including Common Voice, parliamentary meetings, and broadcast corpora
Language model enhancement
Includes a Finnish KenLM 5-gram language model to improve recognition accuracy
Lightweight deployment
Supports 8-bit Adam optimizer, suitable for resource-constrained environments

Model Capabilities

Finnish speech-to-text
Short audio transcription (≤20 seconds)
Speech recognition with language model

Use Cases

Speech transcription
Meeting minutes automation
Convert Finnish parliamentary meeting recordings into text records
WER 5.93% on parliamentary dataset
Voice assistant development
Provide voice interaction foundation for Finnish smart devices
CER 1.40% on Common Voice 9.0
EdTech
Language learning tools
Used for Finnish pronunciation assessment systems
Featured Recommended AI Models
AIbase
Empowering the Future, Your AI Solution Knowledge Base
© 2025AIbase