AIbase
EN
Home
Wavlm Libri Clean 100h Base
W

Wavlm Libri Clean 100h Base

Developed by patrickvonplaten
An automatic speech recognition model fine-tuned on the LIBRISPEECH_ASR - CLEAN dataset based on microsoft/wavlm-base
Speech Recognition
Transformers
#High-precision speech recognition#LibriSpeech optimization#WavLM architecture
Downloads 6,515
Release Time : 3/2/2022

Model Overview

This model is an optimized WavLM base version for English speech recognition tasks, fine-tuned on 100 hours of clean speech data with a low word error rate.

Model Features

Efficient fine-tuning
Fine-tuned on 100 hours of clean speech data, significantly improving the recognition accuracy of the base model
Low word error rate
Achieved a word error rate (WER) of 0.0675 on the evaluation set, demonstrating excellent performance
Multi-GPU training
Utilized 8 GPUs for distributed training, enhancing training efficiency

Model Capabilities

English speech recognition
Continuous speech to text
High-accuracy transcription

Use Cases

Speech transcription
Automatic meeting minutes generation
Automatically convert meeting recordings into text transcripts
Accuracy approximately 93.25% (based on WER 0.0675 calculation)
Podcast content indexing
Generate searchable text content for audio podcasts
Assistive technology
Real-time caption generation
Provide real-time captions for video or live streaming content
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