Fix full-corpus zip layout in croissant.json and stimuli README

The zips store WAVs at the archive root, not under audio_full/. Fix the Croissant FileSet includes and the unzip command.

croissant.json

@@ -491,7 +491,7 @@
       "@type": "cr:FileObject",
       "@id": "audio-zip-01",
       "name": "audio_full_01.zip",
-      "description": "Zip 1 of 14 covering the full TTS-rendered audio corpus. ~1.9 GB. Contains a subset of audio_full/{conv_id}_{turn_no}_{response_idx}.wav files; the per-zip conv_id rollup is in stimuli/main_study/manifest.csv. The full set joins to stimuli/main_study/full_audio_index.csv on audio_file.",
+      "description": "Zip 1 of 14 covering the full TTS-rendered audio corpus. ~1.9 GB. Contains a subset of the {conv_id}_{turn_no}_{response_idx}.wav files at the archive root; the per-zip conv_id rollup is in stimuli/main_study/manifest.csv. The full set joins to stimuli/main_study/full_audio_index.csv on audio_file.",
       "contentUrl": "https://huggingface.co/datasets/sciminds/same-words-different-judgements/resolve/main/stimuli/main_study/audio_full_zips/audio_full_01.zip",
       "encodingFormat": "application/zip",
       "sha256": "cc3384114d0859e904a5541b1717648eac998e381055c75c44ed26e0b0754eca"
@@ -626,7 +626,7 @@
       "@type": "cr:FileSet",
       "@id": "full-audio-wavs",
       "name": "full-audio-wavs",
-      "description": "All 21,603 TTS-rendered .wav files distributed across audio_full_01.zip..audio_full_14.zip. After extracting all 14 zips into a single directory, files are at audio_full/{conv_id}_{turn_no}_{response_idx}.wav and join to the audio_file column in full_audio_index.csv. The per-zip conv_id partition is documented in stimuli/main_study/manifest.csv.",
+      "description": "All 21,603 TTS-rendered .wav files distributed across audio_full_01.zip..audio_full_14.zip. Each zip holds its files at the archive root as {conv_id}_{turn_no}_{response_idx}.wav; extract all 14 into one directory (e.g. stimuli/main_study/audio_full/). Filenames join to the audio_file column in full_audio_index.csv. The per-zip conv_id partition is documented in stimuli/main_study/manifest.csv.",
       "containedIn": [
         {
           "@id": "audio-zip-01"
@@ -672,7 +672,7 @@
         }
       ],
       "encodingFormat": "audio/wav",
-      "includes": "audio_full/*.wav"
+      "includes": "*.wav"
     },
     {
       "@type": "cr:FileObject",

stimuli/main_study/README.md

@@ -16,8 +16,8 @@
   `used_in_study` flag.
 - `audio_full_zips/audio_full_01.zip` … `audio_full_14.zip` — **the
   actual 21,603 .wav files** (16-bit mono 24 kHz PCM, ~26 GB compressed / ~34 GB
-  raw), split across 14 zips of ~1.9 GB each. Filenames inside follow
-  `audio_full/{conv_id}_{turn_no}_{response_idx}.wav` and join to
+  raw), split across 14 zips of ~1.9 GB each. Each zip holds its WAVs at
+  the top level, named `{conv_id}_{turn_no}_{response_idx}.wav`, which join to
   `full_audio_index.csv` on `audio_file`. The per-zip `conv_id`
   partition is in `manifest.csv`.
 
@@ -25,7 +25,7 @@
   ```bash
   hf download sciminds/same-words-different-judgements --repo-type=dataset \
     --include "stimuli/main_study/audio_full_zips/*" --local-dir .
-  for z in stimuli/main_study/audio_full_zips/audio_full_*.zip; do unzip -d stimuli/main_study/ "$z"; done
+  for z in stimuli/main_study/audio_full_zips/audio_full_*.zip; do unzip -q -d stimuli/main_study/audio_full/ "$z"; done
   ```
   After extraction, every filename in `full_audio_index.csv` resolves
   to a file in `stimuli/main_study/audio_full/`.