AIbase
EN
Home
Wavlm Libri Clean 100h Base Plus
W

Wavlm Libri Clean 100h Base Plus

Developed by patrickvonplaten
An automatic speech recognition model fine-tuned on the LIBRISPEECH_ASR - CLEAN dataset based on microsoft/wavlm-base-plus
Speech Recognition
Transformers
#High-precision speech recognition#LibriSpeech optimized#Low word error rate
Downloads 126.17k
Release Time : 3/2/2022

Model Overview

This model is an optimized WavLM model for English speech recognition tasks, fine-tuned on the LibriSpeech clean-100h dataset, achieving a low word error rate (WER).

Model Features

Efficient Fine-tuning
Fine-tuned based on the pre-trained WavLM-base-plus model, fully leveraging the powerful feature extraction capabilities of the pre-trained model
Low Word Error Rate
Achieved a word error rate (WER) of 0.0683 on the evaluation set, demonstrating excellent performance
Multi-GPU Training Optimization
Utilized 8-GPU parallel training with a total batch size of 32, ensuring high training efficiency

Model Capabilities

English speech recognition
Continuous speech-to-text
High-accuracy transcription

Use Cases

Speech Transcription
Audiobook Transcription
Automatically transcribe English audiobook content into text
Achieved a 6.83% word error rate on the LibriSpeech dataset
Meeting Minutes
Automatically convert English meeting recordings into written transcripts
Featured Recommended AI Models
AIbase
Empowering the Future, Your AI Solution Knowledge Base
English简体中文繁體中文にほんご
© 2025AIbase