[Fix] soundclassifier

docs/source/cls/custom_dataset.md

@@ -98,6 +98,8 @@ for epoch in range(1, epochs + 1):
         # Need a padding when lengths of waveforms differ in a batch.
         feats = feature_extractor(waveforms)        
         feats = paddle.transpose(feats, [0, 2, 1])
+        if feats.dim() == 3:
+            feats = feats.unsqueeze(1)
         logits = model(feats)
         loss = criterion(logits, labels)
         loss.backward()

docs/tutorial/cls/cls_tutorial.ipynb

@@ -498,6 +498,8 @@
     "        waveforms, labels = batch\n",
     "        feats = feature_extractor(waveforms)\n",
     "        feats = paddle.transpose(feats, [0, 2, 1])  # [B, N, T] -> [B, T, N]\n",
+    "        if feats.dim() == 3:\n",
+    "            feats = feats.unsqueeze(1)\n",
     "        logits = model(feats)\n",
     "\n",
     "        loss = criterion(logits, labels)\n",
@@ -541,7 +543,9 @@
     "                waveforms, labels = batch\n",
     "                feats = feature_extractor(waveforms)\n",
     "                feats = paddle.transpose(feats, [0, 2, 1])\n",
-    "                \n",
+    "                if feats.dim() == 3:\n",
+    "                    feats = feats.unsqueeze(1)\n",
+    "\n",
     "                logits = model(feats)\n",
     "\n",
     "                preds = paddle.argmax(logits, axis=1)\n",
@@ -576,6 +580,8 @@
     "feats = feature_extractor(paddle.to_tensor(paddle.to_tensor(waveform).unsqueeze(0)))\n",
     "feats = paddle.transpose(feats, [0, 2, 1])  # [B, N, T] -> [B, T, N]\n",
     "print(feats.shape)\n",
+    "if feats.dim() == 3:\n",
+    "    feats = feats.unsqueeze(1)\n",
     "\n",
     "logits = model(feats)\n",
     "probs = nn.functional.softmax(logits, axis=1).numpy()\n",

examples/tess/cls0/local/train.py

@@ -109,6 +109,8 @@ if __name__ == "__main__":
         num_samples = 0
         for batch_idx, batch in enumerate(train_loader):
             feats, labels, length = batch  # feats-->(N, length, n_mels)
+            if feats.dim() == 3:
+                feats = feats.unsqueeze(1)
 
             logits = model(feats)
 
@@ -170,6 +172,9 @@ if __name__ == "__main__":
             with logger.processing('Evaluation on validation dataset'):
                 for batch_idx, batch in enumerate(dev_loader):
                     feats, labels, length = batch
+                    if feats.dim() == 3:
+                        feats = feats.unsqueeze(1)
+
                     logits = model(feats)
 
                     preds = paddle.argmax(logits, axis=1)

paddlespeech/cls/exps/panns/export_model.py

@@ -38,8 +38,9 @@ if __name__ == '__main__':
         model,
         input_spec=[
             paddle.static.InputSpec(
-                shape=[None, None, 64], dtype=paddle.float32)
+                shape=[None, 1, None, 64], dtype=paddle.float32)
-        ])
+        ],
+        full_graph=True)
 
     # Save in static graph model.
     paddle.jit.save(model, os.path.join(args.output_dir, "inference"))

paddlespeech/cls/exps/panns/predict.py

@@ -62,6 +62,9 @@ if __name__ == '__main__':
     model.eval()
 
     feat = extract_features(predicting_conf['audio_file'], **feat_conf)
+    if feat.dim() == 3:
+        feat = feat.unsqueeze(1)
+
     logits = model(feat)
     probs = F.softmax(logits, axis=1).numpy()
 

paddlespeech/cls/exps/panns/train.py

@@ -89,6 +89,8 @@ if __name__ == "__main__":
                 waveforms
             )  # Need a padding when lengths of waveforms differ in a batch.
             feats = paddle.transpose(feats, [0, 2, 1])  # To [N, length, n_mels]
+            if feats.dim() == 3:
+                feats = feats.unsqueeze(1)
 
             logits = model(feats)
 
@@ -150,6 +152,8 @@ if __name__ == "__main__":
                     waveforms, labels = batch
                     feats = feature_extractor(waveforms)
                     feats = paddle.transpose(feats, [0, 2, 1])
+                    if feats.dim() == 3:
+                        feats = feats.unsqueeze(1)
 
                     logits = model(feats)
 

paddlespeech/cls/models/panns/classifier.py

@@ -28,7 +28,9 @@ class SoundClassifier(nn.Layer):
 
     def forward(self, x):
         # x: (batch_size, num_frames, num_melbins) -> (batch_size, 1, num_frames, num_melbins)
-        x = x.unsqueeze(1)
+        if x.dim() == 3:
+            x = x.unsqueeze(1)
+
         x = self.backbone(x)
         x = self.dropout(x)
         logits = self.fc(x)