---
dataset_info:
  features:
  - name: chunk_id
    dtype: string
  - name: audio
    dtype:
      audio:
        sampling_rate: 16000
  - name: transcript_text
    dtype: string
  - name: duration
    dtype: float32
  - name: original_video_id
    dtype: string
  splits:
  - name: train
    num_bytes: 12732333327.84
    num_examples: 23264
  download_size: 12473878794
  dataset_size: 12732333327.84
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
language:
- ar
license: other
tags:
- arabic
- speech
- asr
- tts
- Algerian-arabic
- audio
- paralinguistic
- non-verbal
- algerian
pretty_name: Kahwa Postcast Arabic Speech Dataset
task_categories:
- text-to-speech
- automatic-speech-recognition
size_categories:
- 10K<n<100K
---

# Kahwa Postcast Arabic Speech Dataset

## Dataset Summary

The Kahwa Postcast Arabic Speech Dataset is a large-scale, **first-of-its-kind** Arabic speech corpus containing approximately **110 hours of speech recordings** and corresponding transcripts.

What distinguishes this dataset as a pioneering resource in Arabic language technology is its comprehensive inclusion of **rich non-verbal transcriptions**. Alongside the spoken Arabic text, the transcripts meticulously capture paralinguistic vocalizations, breathing, and emotional cues. This makes it uniquely suited for developing highly expressive, natural-sounding AI models that go beyond flat dictation.

The dataset was created to support Arabic speech technology research and development, including:

* Automatic Speech Recognition (ASR)
* Text-to-Speech (TTS) and **Expressive TTS**
* Speech Foundation Models
* Audio-Text Alignment
* Speaker Adaptation
* Paralinguistic and Emotion Recognition
* Arabic Language Technology Research

Transcriptions were generated and curated by the dataset creator using an AI-assisted transcription pipeline and additional quality-control procedures to ensure the accurate logging of non-verbal tokens.

## Non-Verbal Vocalization Tags

The transcripts include a comprehensive set of non-verbal tokens to capture the true nuance of human speech and breathing. The supported tags are:

`<laugh>` `<cry>` `<weep>` `<sob>` `<scream>` `<shout>` `<whisper>` `<sigh>` `<gasp>` `<groan>` `<moan>` `<pause>` `<hes>` `<stutter>` `<breath>` `<sniff>` `<cough>` `<throat_clear>`

## Research & Application Use Cases

With 110 hours of speech from a single speaker combined with granular non-verbal tagging, the dataset is particularly suitable for:

* **High-fidelity conversational AI:** Training models that understand and generate natural human hesitations, pauses, and breaths.
* **Highly expressive Arabic TTS:** Voice cloning that can synthesize emotion (laughing, whispering, crying) naturally.
* Speaker adaptation and speaker representation learning.
* Long-form ASR training with robust noise/vocalization handling.
* Foundation model pretraining and fine-tuning.
* Research on Arabic speech, narration styles, and paralinguistics.

## Dataset Statistics

| Metric | Value |
| --- | --- |
| Language | Arabic |
| Speaker | Kahwa Postcast |
| Total Audio Duration | ~110 Hours |
| Number of Speakers | 1 |
| Tasks | ASR, TTS, Expressive TTS, Paralinguistics |
| Transcript Generation | AI-assisted, creator-curated |
| Special Features | Rich non-verbal and emotional transcription tags |
| Format | Audio + Text |


يحتفظ المنشئون الأصليون والقنوات المالكة بكافة الحقوق، ولا يتم ادعاء أي ملكية للملفات الصوتية الأصلية أو المحتوى الصوتي الأساسي.