30 days ago
-
Dataset expansion (speaker-sharded): 8669 clips ({'hi': 2669, 'en': 6000})
-
Remove legacy flat audio batch 10
-
Remove legacy flat audio batch 9
-
Remove legacy flat audio batch 8
-
Remove legacy flat audio batch 7
-
Remove legacy flat audio batch 6
-
Remove legacy flat audio batch 5
-
Remove legacy flat audio batch 4
-
Remove legacy flat audio batch 3
-
Remove legacy flat audio batch 2
-
Remove legacy flat audio batch 1
-
Fix: shard audio into speaker subdirectories to bypass Git 10k files limit
-
Add Kaggle 1-click dataset collector notebook for 15H+ data
-
Update: Colab 1-click dataset collector for 15H+ audio
-
Update: fix en_count bug and increase default targets to 15H+ (12k clips)
-
Remove 200 orphan WAVs (no transcript)
-
Remove 200 orphan WAVs (no transcript)
-
Remove 200 orphan WAVs (no transcript)
-
Remove 200 orphan WAVs (no transcript)
-
Remove 200 orphan WAVs (no transcript)
-
Remove 200 orphan WAVs (no transcript)
-
Remove 200 orphan WAVs (no transcript)
-
Remove 200 orphan WAVs (no transcript)
-
Remove 200 orphan WAVs (no transcript)
-
Remove 200 orphan WAVs (no transcript)
-
Remove 200 orphan WAVs (no transcript)
-
Remove 200 orphan WAVs (no transcript)
-
Remove 200 orphan WAVs (no transcript)
-
Remove 200 orphan WAVs (no transcript)
-
Remove 200 orphan WAVs (no transcript)
-
Dataset expansion: 4669 clips ({'hi': 2669, 'en': 2000})
-
Stale-session guard: fresh_download in all/clean modes
-
Clean metadata: 5428 clips, WAVs == rows 1:1
-
Remove 41 orphan WAVs (no transcript)
-
Remove 200 orphan WAVs (no transcript)
-
Remove 200 orphan WAVs (no transcript)
-
Remove 200 orphan WAVs (no transcript)
-
Remove 200 orphan WAVs (no transcript)
-
Remove 200 orphan WAVs (no transcript)
-
Remove 200 orphan WAVs (no transcript)
-
Remove 200 orphan WAVs (no transcript)
-
Remove 200 orphan WAVs (no transcript)
-
Remove 200 orphan WAVs (no transcript)
-
Remove 200 orphan WAVs (no transcript)
-
Remove 200 orphan WAVs (no transcript)
-
Dataset expansion: 7669 clips ({'hi': 2669, 'en': 5000})
-
Flagship-male priority: SPEAKERS.TXT verified male quotas + train.360 fallback + token fix
-
Remove 200 orphan WAVs (no transcript)
-
Remove 200 orphan WAVs (no transcript)
-
Remove 200 orphan WAVs (no transcript)