Tunisian Arabic Speech Recordings with Annotations

A collection of Tunisian dialect audio and corresponding annotations for automatic speech recognition tasks. The dataset was created by Linagora to train their Linto Tunisian dialect STT model, with the first packaged version released in 2023.

Use Cases

• Train an automatic speech recognition model to transcribe Tunisian Arabic audio into text.
• Fine-tune a speech-to-text model on code-switching patterns present in the Tunisian dialect audio.
• Benchmark STT model performance on dialectal Arabic using the provided audio and annotation pairs.
• Analyze phonetic or lexical features specific to Tunisian Arabic from the annotated speech data.

Strengths

• Dataset is described as diverse, containing multiple sources of Tunisian dialect audio.
• Specifically packaged for training a production STT model (linagora/linto-asr-ar-tn).

Limitations

• Specific quantitative details like row count, audio duration, and speaker demographics are unknown.
• Potential for bias towards the specific sources and recording conditions used for collection.

Provenance

Source

Linagora (linagora).

Collection Method

Collected from multiple sources of Tunisian dialect audio and annotated for speech recognition.

Freshness

Last updated on the Hugging Face platform in April 2025.

Geography

Tunisia, focusing on the Tunisian Arabic dialect.