Uyghur speech and transcripts assembled from local WAV/TXT clips and the sources below. The columns are audio, sentence, ref_audio, and ref_text. Both audio columns use embedded mono PCM16 WAV, physically resampled to 16,000 Hz. Transcripts retain their original text, with outer whitespace removed.

Split	Rows	Audio hours
train	112,096	240.71
validation	12,456	26.85

Both original splits of each source were pooled before random row selection (per-repository cap: 6000, seed 42). The selected rows were combined with all 106,142 provided local clips before a fresh split (validation target 10.0%, seed 42). All selected rows are retained. Identical normalized audio stays in the same split, so the final ratio can differ from the target. This is a clip split; speakers/recordings are not held out.

References

Each imported example receives a randomly selected reference from the selected rows of the same source repository and final split, using the build seed. The target waveform and its exact duplicates are excluded. References can be reused by different examples. ref_text is the transcript of ref_audio. Local audiobook rows have ref_audio=null and ref_text="". Imported rows with no different waveform available in their source/split also have blank references and are counted in the build summary. References are grouped by repository; speaker identity is not verified. Reported audio hours count the target audio only, with repeated reference duration reported separately in the summary.

Sources

piyazon/Radio-Ug (6,000 of 60,394 rows; revision 104c3f43cf7c5e38f991cd7f07be5780b7653565; source license metadata: not specified).
piyazon/AliKurban_Hayirhox_Qvxkvnlvk (410 of 410 rows; revision 41cd6b22fa7e50227b35272ae207c1144b276311; source license metadata: not specified).
piyazon/Radio-Girl-Ug (6,000 of 230,695 rows; revision dbbd3a54e0d2bded18804c9de436a671312589d8; source license metadata: not specified).
piyazon/Arhip-Program-Ug (6,000 of 15,850 rows; revision 576936329db5c4b4144f46e7d56213fdf80f34e0; source license metadata: not specified).

This merge does not relicense the source recordings or assign a new blanket license. See the source repositories for their applicable terms.

Access

Access requires a request and manual approval by the dataset owner. The uploader sets this in the repository settings before publishing audio files.

Downloads last month: 23

Total file size:

29.8 GB