W

Wav2vec2 Large Xlsr Marathi

Developed by sumedh
A Marathi automatic speech recognition model fine-tuned from facebook/wav2vec2-large-xlsr-53, trained on OpenSLR Marathi dataset with a test set word error rate of 12.7%.
Downloads 5,159
Release Time : 3/2/2022

Model Overview

This is a model specifically designed for Marathi automatic speech recognition (ASR), fine-tuned from Facebook's wav2vec2-large-xlsr-53 architecture, suitable for 16kHz sampled speech input.

Model Features

Low word error rate
Achieves a word error rate (WER) of 12.7% on the test set, demonstrating excellent performance.
Gender adaptability
Although trained only on female speech data, it performs well on male speech as well.
No language model required
Can be used directly without additional language model support.

Model Capabilities

Marathi speech recognition
16kHz sample rate audio processing

Use Cases

Speech transcription
Marathi speech to text
Convert Marathi speech content into text
12.7% word error rate
Voice assistants
Marathi voice command recognition
Used to develop voice assistants that understand Marathi
Featured Recommended AI Models
AIbase
Empowering the Future, Your AI Solution Knowledge Base
Š 2025AIbase