model + card + eval artifacts

README.md

@@ -28,11 +28,10 @@ audio → 64×101 log-mel spectrogram. Output: one of 12 classes — the keyword
   CC-BY-4.0): 105,829 one-second utterances, 35 words. Official
   validation/testing lists (speaker-disjoint); "unknown" = seeded 10% sample
   of the 25 non-keyword words; "silence" = background-noise crops.
-- **Training recipe**: AdamW lr 3e-3 (cosine-annealed), batch 128, label
-  smoothing 0.1, fp32. Augmentation: ±100 ms time-shift + background-noise
-  mixing (p=0.8, vol U(0,0.1)).
-- **This checkpoint**: 3 epochs on Apple M2 (MPS) — an interim model; a
-  30-epoch Colab T4 run will replace it (this card will be updated).
+- **Training**: 30 epochs on a free Colab T4 (GPU), AdamW lr 3e-3
+  (cosine-annealed), batch 128, label smoothing 0.1, fp32. Best validation
+  accuracy 96.15% at epoch 30. Augmentation: ±100 ms time-shift +
+  background-noise mixing (p=0.8, vol U(0,0.1)).
 - **Features**: log-mel, 64 mels, 25 ms window / 10 ms hop, normalized by
   train-set global mean/std (stored inside the checkpoint).
 
@@ -41,12 +40,15 @@ audio → 64×101 log-mel spectrogram. Output: one of 12 classes — the keyword
 <!-- METRICS_TABLE: produced by evaluate.py, never hand-written -->
 | metric | value |
 |---|---|
-| accuracy | 95.38% |
-| macro-F1 | 95.35% |
-| CPU latency (batch=1, 1 thread, Apple M2) | 1.86 ms mean / 1.96 ms p95 |
+| accuracy | 96.65% |
+| macro-F1 | 96.64% |
+| CPU latency (batch=1, 1 thread, Apple M2) | 1.90 ms mean / 2.08 ms p95 |
 
-Per-class F1 and the confusion matrix: see `metrics.json` and
-`confusion_matrix.png` in this repo.
+Per-class F1 ranges from 0.921 ("unknown", the hardest class) to 0.998
+("silence"); all 10 keywords score ≥0.94. Full per-class table and the
+confusion matrix: see `metrics.json` and `confusion_matrix.png` in this repo.
+Evaluating this checkpoint on the Colab T4 and on an Apple M2 produced
+bit-for-bit identical metrics (reproducible across devices).
 
 ## Usage
 

best.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4f872d71c4fe13745e997121616928efba7cbf01ff6f22b1e4640db1564535d5
-size 504897
+oid sha256:3ec8eed15c9db102f524022beae1455aca2ad46460f665fcf78c40410445c1df
+size 507969

metrics.json

@@ -1,81 +1,81 @@
 {
   "checkpoint": "checkpoints/best.pt",
-  "trained_epochs": 3,
+  "trained_epochs": 30,
   "test_set": "official speech_commands_test_set_v0.02",
   "n_test": 4890,
-  "accuracy": 0.9537832310838446,
-  "macro_f1": 0.9535491997697852,
+  "accuracy": 0.9664621676891616,
+  "macro_f1": 0.9664371919645123,
   "per_class": {
     "silence": {
-      "precision": 0.9645390070921985,
+      "precision": 0.9951219512195122,
       "recall": 1.0,
-      "f1-score": 0.9819494584837545,
+      "f1-score": 0.9975550122249389,
       "support": 408.0
     },
     "unknown": {
-      "precision": 0.8866498740554156,
-      "recall": 0.8627450980392157,
-      "f1-score": 0.8745341614906832,
+      "precision": 0.927860696517413,
+      "recall": 0.9142156862745098,
+      "f1-score": 0.9209876543209876,
       "support": 408.0
     },
     "yes": {
-      "precision": 0.988009592326139,
+      "precision": 0.9951690821256038,
       "recall": 0.9832935560859188,
-      "f1-score": 0.9856459330143541,
+      "f1-score": 0.9891956782713085,
       "support": 419.0
     },
     "no": {
-      "precision": 0.9383886255924171,
-      "recall": 0.9777777777777777,
-      "f1-score": 0.9576783555018138,
+      "precision": 0.9637681159420289,
+      "recall": 0.9851851851851852,
+      "f1-score": 0.9743589743589743,
       "support": 405.0
     },
     "up": {
-      "precision": 0.9437939110070258,
-      "recall": 0.9482352941176471,
-      "f1-score": 0.9460093896713615,
+      "precision": 0.9253393665158371,
+      "recall": 0.9623529411764706,
+      "f1-score": 0.9434832756632064,
       "support": 425.0
     },
     "down": {
-      "precision": 0.9671717171717171,
-      "recall": 0.9433497536945813,
-      "f1-score": 0.9551122194513716,
+      "precision": 0.9747474747474747,
+      "recall": 0.9507389162561576,
+      "f1-score": 0.9625935162094763,
       "support": 406.0
     },
     "left": {
-      "precision": 0.96875,
-      "recall": 0.9781553398058253,
-      "f1-score": 0.9734299516908212,
+      "precision": 0.973621103117506,
+      "recall": 0.9854368932038835,
+      "f1-score": 0.9794933655006032,
       "support": 412.0
     },
     "right": {
-      "precision": 0.9718670076726342,
-      "recall": 0.9595959595959596,
-      "f1-score": 0.96569250317662,
+      "precision": 0.9770992366412213,
+      "recall": 0.9696969696969697,
+      "f1-score": 0.973384030418251,
       "support": 396.0
     },
     "on": {
-      "precision": 0.9516539440203562,
-      "recall": 0.9444444444444444,
-      "f1-score": 0.9480354879594424,
+      "precision": 0.9620253164556962,
+      "recall": 0.9595959595959596,
+      "f1-score": 0.9608091024020228,
       "support": 396.0
     },
     "off": {
-      "precision": 0.94,
-      "recall": 0.9353233830845771,
-      "f1-score": 0.9376558603491272,
+      "precision": 0.9546599496221663,
+      "recall": 0.9427860696517413,
+      "f1-score": 0.9486858573216521,
       "support": 402.0
     },
     "stop": {
-      "precision": 0.9852941176470589,
-      "recall": 0.9781021897810219,
-      "f1-score": 0.9816849816849816,
+      "precision": 0.9854014598540146,
+      "recall": 0.9854014598540146,
+      "f1-score": 0.9854014598540146,
       "support": 411.0
     },
     "go": {
-      "precision": 0.9375,
-      "recall": 0.9328358208955224,
-      "f1-score": 0.9351620947630923,
+      "precision": 0.9649122807017544,
+      "recall": 0.9577114427860697,
+      "f1-score": 0.9612983770287141,
       "support": 402.0
     }
   },
@@ -95,166 +95,166 @@
       0
     ],
     [
+      0,
+      373,
+      1,
       2,
-      352,
-      2,
-      5,
-      6,
-      5,
-      5,
-      9,
-      8,
+      7,
+      3,
+      4,
+      7,
       5,
-      2,
-      7
+      1,
+      1,
+      4
     ],
     [
-      1,
-      4,
-      412,
       0,
+      3,
+      412,
       0,
       0,
-      2,
       0,
+      3,
       0,
       0,
+      1,
       0,
       0
     ],
     [
       0,
+      2,
       0,
+      399,
       0,
-      396,
-      0,
-      3,
-      1,
+      2,
+      2,
       0,
       0,
       0,
       0,
-      5
+      0
     ],
     [
+      1,
       2,
-      7,
       0,
       0,
-      403,
+      409,
       0,
       0,
       0,
-      4,
       6,
-      3,
+      5,
+      2,
       0
     ],
     [
-      1,
-      5,
-      1,
-      8,
       0,
-      383,
+      5,
       0,
+      6,
       2,
+      386,
+      0,
       1,
       0,
       0,
+      1,
       5
     ],
     [
-      1,
-      5,
-      1,
       0,
+      2,
       1,
-      1,
-      403,
       0,
+      2,
+      0,
+      406,
+      1,
       0,
       0,
       0,
       0
     ],
     [
-      1,
-      10,
+      0,
+      7,
+      0,
       0,
       1,
       1,
       0,
-      2,
-      380,
-      0,
+      384,
+      1,
       1,
       0,
-      0
+      1
     ],
     [
+      1,
       3,
-      5,
       0,
       0,
       3,
       0,
-      1,
       0,
-      374,
-      8,
       0,
-      2
+      380,
+      8,
+      1,
+      0
     ],
     [
+      0,
       1,
-      4,
       0,
       0,
-      11,
+      15,
+      1,
       0,
       0,
+      3,
+      379,
       0,
-      5,
-      376,
-      1,
-      4
+      3
     ],
     [
       0,
       2,
       0,
       0,
-      1,
+      0,
       2,
       0,
       0,
       0,
-      2,
-      402,
-      2
+      1,
+      405,
+      1
     ],
     [
-      3,
-      3,
-      1,
-      12,
-      1,
-      2,
+      0,
       2,
       0,
+      7,
+      3,
       1,
       2,
       0,
-      375
+      0,
+      1,
+      1,
+      385
     ]
   ],
   "n_parameters": 119372,
   "weight_file_mb": 0.506619,
   "cpu_latency": {
-    "mean_ms": 1.8594329250845476,
-    "p50_ms": 1.8418545005260967,
-    "p95_ms": 1.9583475965191608,
+    "mean_ms": 1.9022779599981732,
+    "p50_ms": 1.8711250013438985,
+    "p95_ms": 2.0753729018906597,
     "runs": 200,
     "threads": 1
   }