boffire/rradyu-tis-snat

Rradyu Tis Snat — Kabyle Podcasts from Radio Algérie Chaîne 2

Status: work in progress. This README is a first draft with placeholders (marked TODO) to fill in as the dataset grows. Metadata above (license, size_categories) will need updating as the collection is built out.

Dataset Description

This dataset is a collection of Kabyle-language ("Taqbaylit") audio podcasts from Radio Algérie Chaîne 2 (podcast.radioalgerie.dz), the Algerian public radio channel broadcasting mainly in Tamazight (Berber) varieties, with Kabyle as its primary language.

• Language: Kabyle (kab), a Northern Berber (Tamazight/Amazigh) language spoken primarily in Kabylie, Algeria.
• Domain: Radio broadcast speech — news bulletins, talk shows, cultural and religious programs, magazines.
• Modality: Audio only (mp3), no transcripts at this stage.

TODO: expand with your motivation for building this dataset (e.g. ASR pretraining/fine-tuning for Kabyle, language documentation, corpus for a specific research project, etc.)

Dataset Structure

data/
  <show-slug>/
    <episode-title>.mp3
    ...
metadata.csv

Audio files are grouped by show (radio program) under data/<show-slug>/. A single metadata.csv at the repo root follows the AudioFolder convention, so the dataset loads directly with:

from datasets import load_dataset

ds = load_dataset("boffire/rradyu-tis-snat")

Data Fields

Field	Type	Description
file_name	string	Relative path to the audio file (links row to file)
show_id	int	Internal ID of the show on radioalgerie.dz
show_title	string	Title of the radio show/program
episode_title	string	Episode title/date as published on the source site
source_url	string	Original mp3 URL on radioalgerie.dz

TODO: add duration, sampling_rate, channels columns once probing is added to the collection script. TODO: add a split column if/when train/validation/test splits are defined.

Data Splits

TODO: no splits defined yet — currently a single unsplit collection.

Source Data

Collection process

Audio files were scraped directly from the public podcast pages of Radio Algérie Chaîne 2. Shows were filtered heuristically: programs explicitly labeled in another language/Berber variety (Mozabite, Chaoui, Tergui, Arabic, French-language magazines, etc.) were excluded; remaining shows were assumed to be in Kabyle, since Chaîne 2 is historically Algeria's Kabyle-language channel.

Caveat: this is a heuristic, title-based filter, not a per-episode language verification. Some misclassified episodes may remain. TODO: note here if/when episodes get manually or automatically re-verified (e.g. via a language ID model).

Who are the source language producers?

Radio Algérie (Établissement Public de Radiodiffusion Sonore), Chaîne 2 — professional radio hosts, journalists, and guests broadcasting in Kabyle.

Considerations for Using the Data

Licensing / Rights

TODO: this needs to be resolved before any public/production use. The audio is broadcast content from a state-owned Algerian public radio station; its redistribution terms are not yet confirmed. Until clarified, treat this repo as private/research-use-only and do not assume permissive reuse rights.

Known limitations

• Show-level language filtering (see Collection process) may include a small number of non-Kabyle episodes.
• No transcripts, speaker labels, or content moderation have been applied.
• Episodes are long-form (typically 15–30 min); most ASR/fine-tuning pipelines will want to segment these further (e.g. via VAD) — segmentation is left to downstream users rather than baked into this raw archive.
• Audio quality reflects the original radio encoding (mp3); no re-encoding has been performed to avoid further lossy degradation.

Social impact / bias

TODO: note demographic/topic coverage once known (e.g. skew toward news vs. entertainment, presenter gender balance, time period covered, etc.)

Additional Information

• Dataset curators: TODO (your name / organization)
• Contact: TODO
• Source: https://podcast.radioalgerie.dz/fr/podcasts/Chaine-Deux/
• Repository: https://huggingface.co/datasets/boffire/rradyu-tis-snat

Citation

TODO: add a citation block.

@misc{rradyu_tis_snat,
  title  = {Rradyu Tis Snat: Kabyle Podcasts from Radio Alg\'erie Cha\^ine 2},
  author = {TODO},
  year   = {2026},
  url    = {https://huggingface.co/datasets/boffire/rradyu-tis-snat}
}

Changelog

• TODO: log batches/dates as you push (e.g. "2026-07-28: initial batch, N shows, M episodes").