Oriel — AI & technology worth noticing, every day.

bitmanagerai-speech-lab/darbuka_arabic_tts_dataset

Dataset · https://huggingface.co/datasets/bitmanagerai-speech-lab/darbuka_arabic_tts_dataset

Darbuka Arabic Darbuka Arabic brings together Arabic speech, automatic transcripts, and audio-quality measurements for speech recognition and text-to-speech research, with a focus on Egyptian and Saudi speech. At a glance Speech 1,944 hours across 943,731 clips Clip length 2–20 seconds Audio FLAC, original 44.1 kHz or 48 kHz sample rate Transcripts Three ASR outputs and a consensus transcript Size Approximately 885 GB, streamable without a full download… See the full description on the dataset page: https://huggingface.co/datasets/bitmanagerai-speech-lab/darbuka_arabic_tts_dataset.

First seen by Oriel on .

AI summary

Darbuka Arabic is a streamable 1,944-hour dataset of Egyptian and Saudi Arabic speech, transcripts, and quality measurements for speech recognition and text-to-speech research.

  • The collection contains 943,731 FLAC clips lasting 2–20 seconds, with original 44.1 kHz or 48 kHz sampling rates and approximately 885 GB of data.
  • Each clip includes a PriorityROVER consensus transcript, individual Audar, Cohere, and Jais transcripts, ASR agreement metrics, quality measurements, dialect labels, and source information.
  • 98.01% of audio hours have DistillMOS scores of at least 3.0, while 83.59% have ASR consistency of at least 80%.
  • Recordings were sourced under Creative Commons Attribution licenses, segmented using speaker segmentation and speech-end detection, and include measurements covering perceptual quality, music, anti-spoofing, and TTS suitability.
  • The dataset supports streaming through Hugging Face Datasets, with a preview configuration for inspecting a single shard without downloading the full collection.

#arabic speech #text-to-speech #speech recognition #audio datasets

Generated by gpt-5.6-sol. Check the original for details.

Metric trends

Last 30 days · Up to 120 recorded samples per source and metric, with reposts kept separate · First and latest retained

Likes

Total likes reported by the source.

  • Hugging Face Datasets: 22, latest observation ; +7 over observed period since

Downloads (30d)

Rolling 30-day downloads. A change in this value is not new downloads during the observed period.

  • Hugging Face Datasets: 3,384, latest observation ; +2,239 over observed period since

Aliases

SourceExternal IDURL
Hugging Face Datasetsbitmanagerai-speech-lab/darbuka_arabic_tts_datasethttps://huggingface.co/datasets/bitmanagerai-speech-lab/darbuka_arabic_tts_dataset Read discussion