Update README + metadata: severity recomputed from canonical speaker mapping (test Mild 197->421, Severe 438->214).
Browse files- README.md +11 -4
- dataset_metadata.json +46 -5
README.md
CHANGED
|
@@ -79,10 +79,10 @@ corpus×severity confound. All LibriSpeech rows are labelled
|
|
| 79 |
### By severity
|
| 80 |
```
|
| 81 |
{
|
| 82 |
-
"Mild":
|
| 83 |
"Moderate": 585,
|
| 84 |
"Normal": 560,
|
| 85 |
-
"Severe":
|
| 86 |
}
|
| 87 |
```
|
| 88 |
|
|
@@ -131,8 +131,15 @@ doesn't have to remap on the fly:
|
|
| 131 |
prefix prevents speaker-invariance objectives, LSO splits, and per-speaker
|
| 132 |
statistics from silently merging them.
|
| 133 |
|
| 134 |
-
|
| 135 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 136 |
|
| 137 |
## Schema
|
| 138 |
|
|
|
|
| 79 |
### By severity
|
| 80 |
```
|
| 81 |
{
|
| 82 |
+
"Mild": 421,
|
| 83 |
"Moderate": 585,
|
| 84 |
"Normal": 560,
|
| 85 |
+
"Severe": 214
|
| 86 |
}
|
| 87 |
```
|
| 88 |
|
|
|
|
| 131 |
prefix prevents speaker-invariance objectives, LSO splits, and per-speaker
|
| 132 |
statistics from silently merging them.
|
| 133 |
|
| 134 |
+
3. **`severity` is recomputed per speaker from the canonical mapping**
|
| 135 |
+
(TORGO: clinical severity from the TORGO speaker sheet; UA-Speech:
|
| 136 |
+
intelligibility banding → `Very-low (<=19%) -> Severe`,
|
| 137 |
+
`Low (28-39%) / Mid (58-62%) -> Moderate`, `High (>=86%) -> Mild`). This
|
| 138 |
+
corrects a prior UA-Speech labelling bug where `M01` was `Mild` and
|
| 139 |
+
`M09` / `M10` were `Severe`.
|
| 140 |
+
|
| 141 |
+
Other columns (`diagnosis`, `condition`, `intelligibility`) are preserved
|
| 142 |
+
verbatim from their source corpora.
|
| 143 |
|
| 144 |
## Schema
|
| 145 |
|
dataset_metadata.json
CHANGED
|
@@ -17,16 +17,16 @@
|
|
| 17 |
"LibriSpeech": 280
|
| 18 |
},
|
| 19 |
"test_by_severity": {
|
| 20 |
-
"Mild":
|
| 21 |
"Moderate": 585,
|
| 22 |
"Normal": 560,
|
| 23 |
-
"Severe":
|
| 24 |
},
|
| 25 |
"test_allocation_by_severity": {
|
| 26 |
-
"Mild":
|
| 27 |
"Moderate": 585,
|
| 28 |
"Normal": 560,
|
| 29 |
-
"Severe":
|
| 30 |
},
|
| 31 |
"test_by_speaker": {
|
| 32 |
"UA-Speech_M10": 193,
|
|
@@ -57,7 +57,48 @@
|
|
| 57 |
"colliding_ids_fixed_across_TORGO_and_UA_Speech": [
|
| 58 |
"M01",
|
| 59 |
"M05"
|
| 60 |
-
]
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 61 |
},
|
| 62 |
"columns": [
|
| 63 |
"audio",
|
|
|
|
| 17 |
"LibriSpeech": 280
|
| 18 |
},
|
| 19 |
"test_by_severity": {
|
| 20 |
+
"Mild": 421,
|
| 21 |
"Moderate": 585,
|
| 22 |
"Normal": 560,
|
| 23 |
+
"Severe": 214
|
| 24 |
},
|
| 25 |
"test_allocation_by_severity": {
|
| 26 |
+
"Mild": 421,
|
| 27 |
"Moderate": 585,
|
| 28 |
"Normal": 560,
|
| 29 |
+
"Severe": 214
|
| 30 |
},
|
| 31 |
"test_by_speaker": {
|
| 32 |
"UA-Speech_M10": 193,
|
|
|
|
| 57 |
"colliding_ids_fixed_across_TORGO_and_UA_Speech": [
|
| 58 |
"M01",
|
| 59 |
"M05"
|
| 60 |
+
],
|
| 61 |
+
"severity_recomputed_from_canonical_speaker_mapping": true
|
| 62 |
+
},
|
| 63 |
+
"speaker_severity_mapping": {
|
| 64 |
+
"TORGO": {
|
| 65 |
+
"M01": "Severe",
|
| 66 |
+
"M02": "Severe",
|
| 67 |
+
"M03": "Mild",
|
| 68 |
+
"M04": "Severe",
|
| 69 |
+
"M05": "Moderate",
|
| 70 |
+
"F01": "Severe",
|
| 71 |
+
"F03": "Moderate",
|
| 72 |
+
"F04": "Mild",
|
| 73 |
+
"MC01": "Normal",
|
| 74 |
+
"MC02": "Normal",
|
| 75 |
+
"MC03": "Normal",
|
| 76 |
+
"MC04": "Normal",
|
| 77 |
+
"FC01": "Normal",
|
| 78 |
+
"FC02": "Normal",
|
| 79 |
+
"FC03": "Normal"
|
| 80 |
+
},
|
| 81 |
+
"UA-Speech": {
|
| 82 |
+
"M01": "Severe",
|
| 83 |
+
"M04": "Severe",
|
| 84 |
+
"M12": "Severe",
|
| 85 |
+
"F03": "Severe",
|
| 86 |
+
"M05": "Moderate",
|
| 87 |
+
"M06": "Moderate",
|
| 88 |
+
"M07": "Moderate",
|
| 89 |
+
"M11": "Moderate",
|
| 90 |
+
"F02": "Moderate",
|
| 91 |
+
"F04": "Moderate",
|
| 92 |
+
"M03": "Moderate",
|
| 93 |
+
"F01": "Moderate",
|
| 94 |
+
"M08": "Mild",
|
| 95 |
+
"M09": "Mild",
|
| 96 |
+
"M10": "Mild",
|
| 97 |
+
"M14": "Mild",
|
| 98 |
+
"M02": "Mild",
|
| 99 |
+
"F05": "Mild"
|
| 100 |
+
},
|
| 101 |
+
"LibriSpeech": "Normal (all rows)"
|
| 102 |
},
|
| 103 |
"columns": [
|
| 104 |
"audio",
|