Why this matters
Large, studio-quality single-speaker collections in dialectal Arabic are rare. This corpus supplies ~100 hours of Egyptian Arabic narration from one consistent voice with native 24 kHz audio and aligned transcripts—enough data to fine-tune production TTS models rather than only perform few-shot cloning, and valuable for ASR adaptation to Egyptian (Masri) dialect.
What Sets It Apart
- Single-speaker scale: ~100 hours (15,653 clips) recorded across 247 source episodes, concentrating long-form narrated utterances that support long-context prosody learning rather than conversational fragments.
- Production consistency: studio narration on one microphone chain (consistent loudness, low noise), delivered as 24 kHz, 16-bit PCM WAV—no resampling required for modern neural vocoders.
- Dialectal transcripts: machine-generated, normalized Egyptian Arabic (undiacritized, no punctuation), preserving colloquial orthography (e.g. عايز, بقى, ازاي) that differs from MSA corpora.
- Rich metadata: per-clip ASR confidence, source video id, start/end offsets and a promo flag to filter sponsor segments; split assignment is disjoint by source video for robust evaluation.
- Licensing & provenance: manifests/transcripts/segmentation released under CC BY-NC 4.0; underlying source recordings remain with original rights holders—dataset intended for research/non-commercial use.
Who It's For and Trade-offs
Great fit if you need to fine-tune TTS or adapt ASR to Egyptian Arabic using a single, consistent voice (audiobooks, IVR, conversational agents), or to study dialectal phonology and prosody with studio-quality narration. Use the provided confidence/promo fields and timestamps to build stricter training subsets.
Look elsewhere if you need multi-speaker, spontaneous conversational, noisy/telephony, or commercially licensed voice assets: transcripts are machine-generated (expect errors), text is undiacritized (models must learn vowelization), most clips are long-form (~18–19s), and the CC BY-NC 4.0 license prohibits commercial use without further permission. If a guaranteed single-speaker acoustic guarantee is required, run speaker-embedding verification and filter out outliers.
