TuniSpeech-21h: 21-Hour Tunisian Arabic Speech Corpus

TuniSpeech-21h is a 21-hour speech corpus designed for Tunisian Arabic (Derja). It was developed by TuniSpeech-AI to address the underrepresentation of this dialect in Automatic Speech Recognition (ASR). The dataset is compiled from social media and broadcast materials, capturing spontaneous speech and diverse linguistic characteristics.

Use Cases

• Train Automatic Speech Recognition (ASR) models based on the 21 hours of Tunisian Arabic speech.
• Study dialectal variations and spontaneous speech patterns based on the diverse linguistic characteristics mentioned.
• Benchmark speech recognition performance for Tunisian Arabic using the corpus specifically designed for this purpose.
• Develop language models or pronunciation dictionaries for Tunisian Arabic based on the transcribed speech data.

Strengths

• Contains 21 hours of speech data, providing a substantial resource for model training.
• Specifically targets the under-represented Tunisian Arabic (Derja) dialect.
• Compiled from diverse sources including social media (YouTube, Facebook) and broadcast materials, likely capturing a range of speech styles.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count and file formats are unknown, which may limit suitability assessment.
• Last updated 2026-03-31 16:24:53; freshness should be verified.

Provenance

Source

TuniSpeech-AI

Collection Method

Compiled from social media (YouTube and Facebook) and broadcast materials.

Freshness

Last updated 2026-03-31 16:24:53.

Geography

Tunisian Arabic (Derja) dialect region.