Add dataset README

README.md

@@ -1,3 +1,26 @@
----
-license: cc-by-3.0
----
+# ccu-hf-data 
+
+This directory contains speech data, data manifests, license, and attribution metadata for the ccu-hf-data dataset,a multilingual speech corpus derived from YouTube videos collected under Creative Commons
+licensing for non-commercial academic research. It was inspired by data collected for the DARPA CCU [(Cross-Cultural Understanding Program)] (https://www.darpa.mil/research/programs/computational-cultural-understanding) and meant to resemble that data. 
+
+## Attribution
+
+When using or redistributing items from this dataset, please credit each item as:
+
+> *"[title]"* by [uploader] ([uploader_id]), [webpage_url],
+> licensed under Creative Commons Attribution (reuse allowed)
+> *(or: license unconfirmed, selected via YouTube CC filter)*.
+> Audio extracted, segmented, and transcripts cleaned for speech research.
+
+See [ATTRIBUTION.txt](ATTRIBUTION.txt) for full license conditions and per-language breakdown.
+
+## Files
+
+Each language has its own license file and subdirectory of the form `<lang>_<file_type>/` with:
+
+- `license_<lang>_filtered.csv`: per-video metadata and license status
+- <lang> : zh, he, ko, tr
+- <file_type> : train, test, dev
+
+The contents of each subdirectory contains a tsv-based manifest containing the file name and the transcript of each segment, and a sub-sub-directory (`files/`) containing the actual file segments. 
+