abdelhaqueidali/Amazigh-Lyrics-Collection

Dataset Card for Amazigh Lyrics Dataset

This dataset is a curated collection of Amazigh music lyrics transcribed and written in the Amazigh script (Tifinagh) using standard orthography. It serves as an open resource to address the acute scarcity of structured text data types for modern Amazigh digital applications.

Dataset Details

Dataset Description

The Amazigh Lyrics Dataset is a work-in-progress (WIP) corpus designed to consolidate musical oral tradition into a standardized digital format. The audio metadata fields (title, artist, album) are provided directly in Amazigh.

To prioritize linguistic uniformity and digital utility, the data is made as standard in its orthography as possible for all the songs included in the dataset. The texts lean heavily toward standard spelling rules rather than transcribing exact local phonetic variants. While the dataset targets a diverse set of dialects, the content primarily features music from Morocco, deeply concentrated around the Southwestern region (Tashelhit), from Sous and Asamer.

• Curated by: Abdelhaque Id Ali
• Language(s) (NLP): Amazigh (Standard Moroccan Amazigh, written in the Amazigh script)
• License: Open for research and educational purposes (Fair Use / Copyright belongs to original owners)

Uses

Direct Use

• Training and fine-tuning Language Models (LLMs) on standard Tifinagh text structures.
• Automatic Speech Recognition (ASR) alignment when combined with external audio inputs.
• Linguistic research into Amazigh syntax, poetic vocabulary, and cross-dialectal standardization.

Out-of-Scope Use

Commercial redistribution or commercialization of raw lyrics without explicit clearance from the copyright holders.

Dataset Structure

The dataset currently exports as a table containing the following fields:

• title: The song title written in Amazigh.

• artist: The recording artist/band name.

• album: The album title (where available). The labeled songs here with "AI" are songs made using song generation tools by their creators.

• lyrics: The cleaned, plain-text lyrics in Tifinagh script, stripped of temporal metadata.

• Language codes explain:

• shi-sous_1: Sous and Atlas who speak with gh (ɣ/ⵖ)

• shi-sous_2: Sous and Atlas who speak with kh (x/ⵅ)

• shi-sous_3: Sous who speak with H (ḥ/ⵃ)

• shi-asamer: Asamer who speak with /ḵ/ and /ḡ/

• tzm: Central Atlas which differ so much from Shi

• rif: Riffian which differ from Central Atlas

• tmh: All of the Toareg dialects

• kab: Kabyle dialect

these codes are not meant to specify the variant accurate, they are designed to keep track of the big differences regarding sounds dialects in a continuum way, so a Central Atlas Amazigh song that may sound like Sous Amazigh dialect might be labeled shi-sous. When the variant is obvious, it gets its correct lang code.

Dataset Creation

Curation Rationale

The Amazigh language remains highly under-resourced in the digital space. Text assets—especially dynamic cultural items like music lyrics—are heavily scarce and fragmented across online spaces. This project aims to provide high-quality, normalized Tifinagh data to empower developers and researchers.

Source Data

Data Collection and Processing

The lyrics in this corpus are created and written by the dataset author in the Amazigh script (Tifinagh) with higher accuracy of the standard orthography, or developed by adapting and processing existing written forms across Latin, Arabic, or Amazigh scripts. Timestamps and system bracket tags (such as [...] or <...>) are stripped during extraction to yield structured, plain-text columns.

Future Roadmap

• Streaming Integration: Future iterations will incorporate direct Spotify and YouTube URLs to bridge the text entries back to their original metadata sources and streaming platforms.
• Time Synchronization: Inclusion of precise, line-by-line .lrc timestamp data for synchronization workflows.
• Linguistic Tagging: Addition of language and localized dialect tags will be added in the future, while preserving the centralized standardized orthographic text structure.

Bias, Risks, and Limitations

• Dialectal Balance: The text collection is heavily skewed toward Moroccan varieties (Sous and Asamer regions). It may not fully represent Northern Moroccan (Rifian) or broader Amazigh varieties at this current stage.
• Standardization over Phonetics: Because the dataset favors standard spelling conventions over raw auditory phonetic variations, researchers focusing strictly on exact localized phonetic acoustics should use it with caution.

Copyright & Takedown Policy

No copyright infringement is intended by this work. The full rights to all underlying lyrical compositions belong strictly to their respective authors and copyright holders. This dataset is compiled solely for educational preservation, language resource development, and digital standardization.

If you are a copyright holder or a representative of one and wish to have your material omitted or modified, please reach out via the contact information provided below to submit a removal request, and your content will be promptly excluded.

Dataset Card Contact

• For Contacting: Abdelhaque Id Ali
• Email: abdelhaqueidali@gmail.com