Datasets:

alakxender
/

dhivehi-asr-errors

dataset_info:
  features:
    - name: truth
      dtype: string
    - name: error
      dtype: string
  splits:
    - name: train
      num_bytes: 360120551
      num_examples: 793692
  download_size: 57539588
  dataset_size: 360120551
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
license: apache-2.0
language:
  - dv
tags:
  - asr_errors
  - dhivehi
pretty_name: 'asr_dhivehi_error '
size_categories:
  - 10K<n<100K

Dhivehi ASR Errors Dataset

Dataset Description

This dataset contains Dhivehi (Maldivian) text with corresponding ASR (Automatic Speech Recognition) errors. It can be used for training error correction models, evaluating ASR systems for Dhivehi, or for linguistic analysis of common error patterns.

Dataset Structure

truth: The original, correct Dhivehi text
error: The text with ASR errors

Usage

from datasets import load_dataset

dataset = load_dataset("alakxender/dhivehi-asr-errors")

Dataset Creation

This dataset was created by combining real ASR errors with synthetically generated errors that mimic common patterns found in ASR systems for the Dhivehi language. The synthetic errors were generated by analyzing patterns in real ASR errors and applying similar transformations, including:

Character substitutions (similar-looking Dhivehi characters)
Diacritic modifications or omissions
Word boundary changes (merging or splitting words)
Character insertions and deletions
Phonetically plausible errors

Example Uses

This dataset can be used for:

Training ASR post-correction models
Analyzing common error patterns in Dhivehi ASR systems
Developing more robust NLP systems for Dhivehi
Linguistic analysis of phonetic similarities in Dhivehi