metadata

language:
  - ar
  - zgh
  - shi
task_categories:
  - translation
pretty_name: Tamazight-Arabic Speech Translation Dataset
size_categories:
  - 10K<n<100K
tags:
  - speech
  - tamazight
  - arabic
  - speech-to-text
  - low-resource
  - north-africa

Tamazight-Arabic Speech Recognition Dataset

This is the Tamazight-NLP organization-hosted version of the Tamazight-Arabic Speech Recognition Dataset. This dataset contains ~15.5 hours of Tamazight (Tachelhit dialect) speech paired with Arabic transcriptions, designed for automatic speech recognition (ASR) and speech-to-text translation tasks.

Dataset Details

Total Examples: 20,344 audio segments
Training Set: 18,309 examples (~8.9GB)
Test Set: 2,035 examples (~992MB)
Total Duration: ~15.5 hours
Audio Format: WAV
Text Format: Arabic script

Quick Start

from datasets import load_dataset
dataset = load_dataset("Tamazight-NLP/Tamazight-Speech-to-Arabic-Text")
for example in dataset["train"]:
    audio, text = example["audio"], example["text"]

Fields

audio: Path, waveform, and sampling rate
text: Arabic transcription
metadata: Duration, timestamps, language, source

Citation

@dataset{tamazight_asr_2024,
  author       = {Contributors},
  title        = {Tamazight-Arabic Speech Recognition Dataset},
  year         = {2025},
  publisher    = {Hugging Face},
  howpublished = {\url{https://huggingface.co/datasets/Tamazight-NLP/Tamazight-Speech-to-Arabic-Text}}
}

Acknowledgments

This dataset was initially created by Soufiane Dahimi