Add Real Audio Bench metadata: README.md

README.md

@@ -0,0 +1,54 @@
+---
+license: other
+task_categories:
+- audio-classification
+- question-answering
+language:
+- en
+pretty_name: Real Audio Bench
+---
+
+# Real Audio Bench
+
+Real Audio Bench is a compact human-recorded benchmark for streaming spoken reasoning. The benchmark contains 186 English audio recordings. Each item is a natural spoken question in which an early answer can be misleading until a later cue is heard.
+
+Audio files are hosted at: https://huggingface.co/datasets/Oulasong/real-audio-bench-v2
+
+## Files
+
+- `audio/`: 186 WAV files.
+- `real_audio_bench_manifest.jsonl`: one metadata row per audio file.
+- `real_audio_bench_manifest.csv`: CSV version of the same manifest.
+- `real_audio_bench_manifest.md`: human-readable table for quick inspection.
+- `answer_corrections_from_reader_packet.json`: audit record of manually corrected answers extracted from the annotated reader packet.
+
+## Manifest schema
+
+Each row contains:
+
+- `audio_path`: relative path of the WAV file in the Hugging Face dataset.
+- `reader`: speaker/reader id.
+- `item`: item id within the reader packet.
+- `source_id`: source item identifier used during benchmark construction.
+- `question`: spoken prompt text.
+- `gold_answer`: corrected reference answer.
+- `gold_answer_original`: original packet answer before manual correction, when applicable.
+- `answer_corrected_from_pdf`: whether the answer was corrected from the annotated reader packet.
+- `category`, `difficulty`, `scenario`, `mechanism`, `duration`: construction metadata.
+- `critical_cue`, `why_early_answer_is_wrong`, `speaker_note`: explanatory metadata from the reader packet.
+
+## Construction summary
+
+The benchmark was constructed from scripted spoken-reasoning prompts and recorded by five internal readers. The final public split keeps 186 recordings after removing ambiguous or unsuitable items from the 200-item reader packet. The reference answers in this release include manual corrections from the annotated packet file.
+
+## Intended use
+
+This benchmark is intended as a compact transfer check for models that answer spoken questions from real human recordings. It is not intended as a large-scale user study or as a general-purpose speech recognition benchmark.
+
+## License
+
+The release license should be confirmed by the dataset owner before publication.
+
+## Release notes
+
+This release contains 186 manifest rows. 23 rows use manually corrected answers from the annotated reader packet.