Create README.md

README.md

@@ -0,0 +1,35 @@
+---
+license: mit
+task_categories:
+- text-generation
+language:
+- en
+tags:
+- agent
+size_categories:
+- 1K<n<10K
+---
+
+# Dataset Card for SAP-Hypo5
+
+<!-- Provide a quick summary of the dataset. -->
+
+SAP-Hypo5 is an open benchmark for LLM-Agent ASR hypothesis correction on dysarthric speech.
+
+### Dataset Description
+
+<!-- Provide a longer summary of what this dataset is. -->
+Following [HyPoradise](https://proceedings.neurips.cc/paper_files/paper/2023/hash/6492267465a7ac507be1f9fd1174e78d-Abstract-Datasets_and_Benchmarks.html), each selected SAP utterance is paired with its reference transcript and the top-5 ASR hypotheses from Whisper-large-v2 fine-tuned on SAP (PD-only challenge release).
+
+### Dataset Sources
+
+<!-- Provide the basic links for the dataset. -->
+
+- **Repository:** https://github.com/xiuwenz2/SAP-Hypo5
+- **Paper:** Towards Robust Dysarthric Speech Recognition: LLM-Agent Post-ASR Correction Beyond WER (Under Review)
+
+### Source Data
+
+<!-- This section describes the source data (e.g. news text and headlines, social media posts, translated sentences, ...). -->
+SAP-Hypo5 is derived from the [SAP](https://speechaccessibilityproject.beckman.illinois.edu/conduct-research-through-the-project) corpus of dysarthric speech. For each SAP utterance, we provide (i) the reference transcript from SAP and (ii) the top-5 ASR hypotheses produced by an open-source ASR model.
+To get the SAP dataset, please sign the [data use agreement](https://speechaccessibilityproject.beckman.illinois.edu/docs/librariesprovider8/default-document-library/data-transfer-and-use-agreement-speech-accessibility-project.pdf?sfvrsn=909bd90f_20) and submit a one-page proposal to speechaccessibility@beckman.illinois.edu.