R

Romanian Wav2vec2

Developed by gigant
A Romanian speech recognition model fine-tuned based on facebook/wav2vec2-xls-r-300m, trained on Common Voice 8.0 and Romanian speech synthesis datasets, ranked first in Romanian speech recognition in the HuggingFace Robust Speech Challenge.
Downloads 88.90k
Release Time : 3/2/2022

Model Overview

This model is used for Romanian speech recognition from 16kHz sampled audio clips, with predicted text in lowercase and without punctuation.

Model Features

High-performance Romanian recognition
Achieved excellent performance with WER 11.73 (CER 2.93) on the Common Voice 8.0 test set
Language model enhancement
Integrated 5-gram language model, significantly improving recognition accuracy (WER reduced from 46.99 to 38.63)
Multi-dataset training
Combined training on Common Voice 8.0 and Romanian speech synthesis datasets
Competition-winning model
Ranked first in Romanian speech recognition in the HuggingFace Robust Speech Challenge

Model Capabilities

Romanian speech recognition
16kHz audio processing
Punctuation-free text output

Use Cases

Speech-to-text
Romanian speech transcription
Convert Romanian speech to text
WER 11.73 on the Common Voice 8.0 test set
Voice assistants
Romanian voice command recognition
Used for front-end speech recognition in Romanian voice assistants
Featured Recommended AI Models
AIbase
Empowering the Future, Your AI Solution Knowledge Base
© 2025AIbase