Dataset Information
- Corpus:
hakkatv_hanzawa - Status:
Public (Gated) - Split Policy:
trainonly (in_ood_testcolumn preserved for traceability without injecting noisy pseudo-labels into eval/test metrics)
Subset Stats
| Subset | Train (Samples / Hours) | CER Before -> After | Eval (Samples / Hours) | Description |
|---|---|---|---|---|
clean_asr |
7,485 / 3.85 | 14.19% -> 9.22% | 0 / 0.00 | Recommended ASR training pool (clean + clean_colloquial, zero overlapping intervals) |
clean |
5,950 / 3.01 | 11.26% -> 5.25% | 0 / 0.00 | Strict gold subset (clean_exact + clean_revised, CER & IPA-CTC verified) |
clean_exact |
5,770 / 2.91 | 10.49% -> 5.23% | 0 / 0.00 | Verbatim original reference matches (text_edited=false) |
clean_revised |
180 / 0.09 | 32.37% -> 5.96% | 0 / 0.00 | Kaldi Biased-LM + Smith-Waterman revised/trimmed utterances (text_edited=true) |
clean_colloquial |
1,535 / 0.84 | 24.08% -> 22.52% | 0 / 0.00 | High CTC+LID colloquial utterances promoted from review_text |
review_text |
3,951 / 2.06 | 29.88% -> 29.27% | 0 / 0.00 | Audio QC passed, text flagged for human review (includes clean_colloquial) |
rejected |
2,944 / 1.25 | 56.81% -> 65.04% | 0 / 0.00 | Filtered out (noise, bleed conflict, Mandarin mismatch, alignment failure) |
all |
12,845 / 6.31 | 26.34% -> 24.97% | 0 / 0.00 | Complete processed corpus across all buckets (100% of utterances) |
Quick Load
from datasets import load_dataset
ds = load_dataset("formospeech/hakkatv_hanzawa_cleaned", "clean_asr", split="train")
ds_revised = load_dataset("formospeech/hakkatv_hanzawa_cleaned", "clean_revised", split="train")
- Downloads last month
- 276