W

Whisper Large V2 Ko

Developed by byoussef
Korean automatic speech recognition (ASR) model fine-tuned based on OpenAI Whisper-large-v2, excelling on Korean datasets
Downloads 94
Release Time : 3/10/2023

Model Overview

This model is a Korean fine-tuned version of OpenAI Whisper-large-v2, specifically optimized for Korean speech recognition tasks, achieving a 2.9% word error rate (WER) on the Zeroth Korean dataset

Model Features

Low word error rate
Only 2.9% word error rate on Korean test sets, demonstrating excellent performance
Multi-GPU training
Efficient training using 7 GPUs with a total training batch size of 224
Optimized training process
Adopted linear learning rate scheduling with 500 warm-up steps, achieving optimal results after 50 epochs

Model Capabilities

Korean speech recognition
Speech-to-text
High-accuracy transcription

Use Cases

Speech transcription
Korean meeting minutes
Automatically transcribe Korean meeting recordings into text
Highly accurate written records
Korean voice assistant
Provide speech recognition capabilities for Korean voice assistants
Accurate voice command recognition
Education
Korean learning applications
Help Korean learners check pronunciation accuracy
Provide accurate pronunciation feedback
Featured Recommended AI Models
AIbase
Empowering the Future, Your AI Solution Knowledge Base
Š 2025AIbase