Add dev/test split tag to sample_id and audio.path

README.md

@@ -57,11 +57,11 @@ size_categories:
 
 Blind **development** subset of the VMC 2026 Track 1 (URGENT 2026 Speech Quality Assessment) data. Each row provides a `sample_id` and audio only -- no labels, no system identifiers, no metadata. Use this set to develop and validate your predictor; the held-out evaluation set lives at [`urgent-challenge/vmc2026-track1-test`](https://huggingface.co/datasets/urgent-challenge/vmc2026-track1-test).
 
-The data is organized into two configs corresponding to two subjective evaluation paradigms: absolute rating (`acr`) and pairwise comparison (`ccr`). The `sample_id` values are namespaced strings such as `vmc2026-track1-acr_489` and `vmc2026-track1-ccr_7233`, so a participant can submit one mixed prediction file and the organizer can split it by the `acr` / `ccr` segment.
+The data is organized into two configs corresponding to two subjective evaluation paradigms: absolute rating (`acr`) and pairwise comparison (`ccr`). The `sample_id` values are namespaced strings such as `vmc2026-track1-dev-acr_489` and `vmc2026-track1-dev-ccr_7233`. The namespace encodes track / split / task: `vmc2026-track1-{dev|test}-{acr|ccr}_{index}`, so a participant can submit one mixed prediction file and the organizer can split it by the `acr` / `ccr` segment, while the `dev` / `test` segment makes the IDs disjoint from the held-out evaluation set.
 
 The dev and test sets are constructed from disjoint base utterances and disjoint speakers (no base noisy clip and no speaker appears in both); for CCR this means no preference-pair audio appears in both sets. Splits are balanced per language at roughly 20% / 80%.
 
-> Note: `sample_id` values are drawn from a single 0..N-1 numbering shared with the test set, so the IDs in this dev set are not contiguous (e.g. `vmc2026-track1-acr_28`, `vmc2026-track1-acr_29`, `vmc2026-track1-acr_30`, `vmc2026-track1-acr_31`, `vmc2026-track1-acr_52`, ...).
+> Note: `sample_id` indices are drawn from a single 0..N-1 numbering shared with the test set, so the IDs in this dev set are not contiguous (e.g. `vmc2026-track1-dev-acr_28`, `vmc2026-track1-dev-acr_29`, `vmc2026-track1-dev-acr_30`, `vmc2026-track1-dev-acr_31`, `vmc2026-track1-dev-acr_52`, ...).
 
 ## `acr` -- Absolute Category Rating
 
@@ -69,7 +69,7 @@ The dev and test sets are constructed from disjoint base utterances and disjoint
 
 | Column | Type | Description |
 |--------|------|-------------|
-| `sample_id` | string | Namespaced identifier such as `vmc2026-track1-acr_489` |
+| `sample_id` | string | Namespaced identifier such as `vmc2026-track1-dev-acr_489` |
 | `audio` | Audio | Speech audio (FLAC, mono); inner `path` is `<sample_id>.flac` |
 | `sample_rate` | int | Sample rate in Hz |
 | `duration` | float | Duration in seconds |
@@ -80,7 +80,7 @@ The dev and test sets are constructed from disjoint base utterances and disjoint
 
 | Column | Type | Description |
 |--------|------|-------------|
-| `sample_id` | string | Namespaced identifier such as `vmc2026-track1-ccr_7233` |
+| `sample_id` | string | Namespaced identifier such as `vmc2026-track1-dev-ccr_7233` |
 | `audio_a` | Audio | Speech audio from system A (FLAC, mono); inner `path` is `<sample_id>_a.flac` |
 | `audio_b` | Audio | Speech audio from system B (FLAC, mono); inner `path` is `<sample_id>_b.flac` |
 | `sample_rate` | int | Sample rate in Hz |
@@ -91,8 +91,8 @@ The dev and test sets are constructed from disjoint base utterances and disjoint
 Submit one space-delimited, headerless `predictions.scp` file with one prediction per line:
 
 ```text
-vmc2026-track1-acr_489 3.42
-vmc2026-track1-ccr_7233 -0.15
+vmc2026-track1-dev-acr_489 3.42
+vmc2026-track1-dev-ccr_7233 -0.15
 ```
 
 The file should contain exactly 1,008 ACR rows and 2,520 CCR rows -- one prediction per `sample_id` in this dataset. ACR scores must lie in [1, 5]; CCR scores in [-3, +3]. Preference accuracy for CCR is derived from the sign of the predicted CMOS.
@@ -111,9 +111,9 @@ Actual contents of the first row of each config (audio arrays truncated for disp
 ```python
 >>> acr[0]
 {
-    'sample_id': 'vmc2026-track1-acr_489',
+    'sample_id': 'vmc2026-track1-dev-acr_489',
     'audio': {
-        'path': 'vmc2026-track1-acr_489.flac',
+        'path': 'vmc2026-track1-dev-acr_489.flac',
         'array': array([-0.00039673, -0.00054932, -0.00045776, -0.00048828, ...]),
         'sampling_rate': 32000,
     },
@@ -123,14 +123,14 @@ Actual contents of the first row of each config (audio arrays truncated for disp
 
 >>> ccr[0]
 {
-    'sample_id': 'vmc2026-track1-ccr_7233',
+    'sample_id': 'vmc2026-track1-dev-ccr_7233',
     'audio_a': {
-        'path': 'vmc2026-track1-ccr_7233_a.flac',
+        'path': 'vmc2026-track1-dev-ccr_7233_a.flac',
         'array': array([0.00054932, 0.00079346, 0.00079346, 0.00076294, ...]),
         'sampling_rate': 16000,
     },
     'audio_b': {
-        'path': 'vmc2026-track1-ccr_7233_b.flac',
+        'path': 'vmc2026-track1-dev-ccr_7233_b.flac',
         'array': array([0.03192139, 0.06243896, 0.07189941, 0.08477783, ...]),
         'sampling_rate': 16000,
     },

data/acr/dev.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f77d4c386ae7f0057b010555dcfb1389552ccfee712a45955653035a392f51ee
-size 111842124
+oid sha256:44a1be607d85c8a3724b3cc9affb0e36daf704908e8eb42c9906ec263219734a
+size 111843319

data/ccr/dev.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:86f18f82fdfb7ba9f9fb2d0c498fced9c18b1ecbb157cbdab2627aa05f19d60c
-size 538595065
+oid sha256:f81a722e1192754f5963a2cb61053ab3b60e6345af39a4781f78a1b51cc0e183
+size 462571134