MoulSot-Full: 1,500 Hours of Moroccan Darija Speech Audio

1,500 hours of Moroccan Darija speech audio were collected from publicly available YouTube content across 51 diverse channels. A high-quality subset of approximately 80 hours has been carefully curated and transcribed. The dataset was created by atlasia and last updated on Hugging Face in May 2026.

Use Cases

• Train automatic speech recognition (ASR) models based on the 80 hours of transcribed Moroccan Darija audio.
• Develop language models for Moroccan Darija based on the transcribed speech content.
• Study dialectal variations and code-switching in North African Arabic based on real-world YouTube content.
• Benchmark speech synthesis models for under-resourced languages using the curated audio subset.

Strengths

• Large-scale corpus of 1,500 hours of raw speech audio.
• Includes a curated, high-quality subset of approximately 80 hours with transcriptions.
• Sourced from 51 diverse YouTube channels, capturing real-world linguistic variety.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• The full 1,500-hour corpus is not transcribed, limiting immediate use for supervised tasks.
• Data may reflect geographic and source bias inherent to the specific YouTube channels used.

Provenance

Source

Publicly available YouTube content across 51 channels (vlogs, podcasts, interviews, commentary).

Collection Method

Built entirely from YouTube content, with a subset curated and transcribed.

Freshness

Last updated 2026-05-05 12:42:46; freshness should be verified.

Geography

Morocco (Moroccan Darija dialect).