CoolFace
Datasetpublic

abdelhaqueidali/Zouhir-Amazigh-Subtitles

Dataset Card for Zouhir-Amazigh-Subtitles This dataset provides parallel sentence-level translations and precise audio timestamps extracted from the YouTube channel of Zouhir Amazigh. It is curated to support Automatic Speech Recognition (ASR), machine translation, and text generation tasks for the Amazigh language. Dataset Details Dataset Summary The dataset contains aligned speech segments, text timestamps, and sentence-level parallel data scraped… See the full description on the dataset page: https://huggingface.co/datasets/abdelhaqueidali/Zouhir-Amazigh-Subtitles.

sourceHugging Faceapache-2.0updated 2mo agoView on Hugging Face
1likes14downloads
Dataset Card

Dataset Card for Zouhir-Amazigh-Subtitles

This dataset provides parallel sentence-level translations and precise audio timestamps extracted from the YouTube channel of Zouhir Amazigh. It is curated to support Automatic Speech Recognition (ASR), machine translation, and text generation tasks for the Amazigh language.

Dataset Details

Dataset Description

  • —Curator: Abdelhaque id ali
  • —Source Material: Zouhir Dehbi Youtube Channel
  • —Language(s): Amazigh, Arabic, and English
  • —License: Apache 2.0

Dataset Summary

The dataset contains aligned speech segments, text timestamps, and sentence-level parallel data scraped from subtitles across videos on Zouhir Amazigh's channel. It serves as a localized resource for fine-tuning speech-to-text models and low-resource translation systems.

Dataset Fields

  • —subtitle_file_name (string): The Youtube ID, the filename of the source media or subtitle file (e.g., VROdu53S_1k.webm is the filename, VROdu53S_1k is the youtube video id), .
  • —line (int64): The sequential index of the subtitle line within the file.
  • —timestamp (string): The start and end timestamps in HH:MM:SS.mmm --> HH:MM:SS.mmm format.
  • —zgh (string): The subtitle text in Standard Moroccan Amazigh (Tifinagh script).
  • —ar (string): The parallel subtitle text in Arabic.
  • —en (string): The parallel subtitle text in English.

Supported Tasks

  • —translation: Training parallel models to convert between Amazigh and target languages.
  • —text-generation: Language modeling and transcript refinement.
  • —text-classification: Categorizing dialogue, topics, or structural syntax.

Dataset Structure

Dataset Creation

Curation Rationale

To address the shortage of speech-aligned datasets for low-resource languages, this repository provides structured, high-quality human-subtitled data ideal for mapping spoken audio to text.

Source Data

  • —Data Source: Subtitles scraped from the @zouhiramazigh138 YouTube channel.
  • —Acknowledgment: Special thanks to Zouhir Dehbi for his dedicated effort in creating high-quality Amazigh subtitles across his video content, which served as the foundation for this dataset.