J

Japanese Wav2vec2 Large Rs35kh

Developed by reazon-research
A Japanese automatic speech recognition model fine-tuned on the large-scale Japanese ASR corpus ReazonSpeech v2.0, based on the wav2vec 2.0 Large architecture
Downloads 244
Release Time : 11/29/2024

Model Overview

This is a high-performance Japanese automatic speech recognition (ASR) model, specifically optimized for Japanese speech recognition tasks, featuring a low character error rate and excellent long speech recognition capability.

Model Features

High-performance Japanese Recognition
Outstanding performance on multiple test sets with an average character error rate (CER) of only 16.25%
Long Speech Processing Capability
Specifically optimized for long speech recognition performance, achieving a CER of only 30.98% on the JSUT-BOOK test set
Trained on Large-scale Dataset
Fine-tuned on the ReazonSpeech v2.0 large-scale Japanese ASR corpus
Supports bfloat16 and Flash Attention
Supports bfloat16 data type and Flash Attention 2 optimization to improve inference efficiency

Model Capabilities

Japanese Speech Recognition
Long Speech Processing
Real-time Speech-to-Text

Use Cases

Speech-to-Text
Japanese Meeting Minutes
Automatically convert Japanese meeting recordings into text transcripts
Average character error rate of 16.25%
Japanese Podcast Transcription
Transcribe Japanese podcast content into text
Long speech recognition CER of 30.98%
Voice Assistant
Japanese Voice Command Recognition
Used for voice command recognition in Japanese voice assistants or smart devices
Featured Recommended AI Models
AIbase
Empowering the Future, Your AI Solution Knowledge Base
© 2025AIbase