Upload README.md with huggingface_hub

README.md

@@ -1,15 +1,21 @@
 ---
-license: cc-by-4.0
-task_categories: [automatic-speech-recognition]
-language: [en]
-pretty_name: Diffusion-ASR hardest 100 test-clean clips (with audio)
+license: apache-2.0
+tags: [automatic-speech-recognition, librispeech, error-analysis]
 ---
-# The 100 test-clean clips with the most errors (diffusion-ASR, 4.60% system) - WITH AUDIO
-
-Listen directly in the viewer: each row carries the LibriSpeech audio (`audio` column),
-the reference (`ref`), what the model wrote (`hyp`), and per-clip error accounting.
-Ranked worst-first (`rank`): these 100 clips (of 2,620) carry 547 errors, 22.4% of the
-system's total. Composition (measured): long sentences (median 37 words) dense with rare
-words and names (12.8% out-of-dictionary vs 9.5% corpus-wide); 34 distinct narrators.
-Full decode set (no audio): [diffusion-asr-testclean-decodes](https://huggingface.co/datasets/Tachyeon/diffusion-asr-testclean-decodes).
-Audio (c) LibriSpeech, Panayotov et al., CC BY 4.0 - redistributed with attribution.
+
+# Worst-100 test-clean clips — audio, transcripts, and the vocabulary finding
+
+The 100 LibriSpeech test-clean clips where the block-4 production model (4.60% WER) made
+the most word errors — with audio embedded so the failures can be *listened to*, plus the
+model's transcript next to the reference for each clip.
+
+## The finding this dataset produced
+
+**47% of the word errors in these clips are on words that never appeared in the 30-hour
+training vocabulary at all** (20,066 distinct words seen). The model was not mishearing
+these words — it had never met them. Repeat offenders are book-specific names (rodolfo,
+montrose, boolooroo). Full breakdown in `worst100_oov_analysis.json`.
+
+This measurement is the primary justification for the 100-hour training phase: a coverage
+problem calls for coverage. See `Diffusion-ASR/pipeline-code-and-docs` for the full recipe
+and evidence record.