add files

funasr_local/export/test/test_onnx.py

@@ -0,0 +1,20 @@
+import onnxruntime
+import numpy as np
+
+
+if __name__ == '__main__':
+    onnx_path = "/mnt/workspace/export/damo/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-pytorch/model.onnx"
+    sess = onnxruntime.InferenceSession(onnx_path)
+    input_name = [nd.name for nd in sess.get_inputs()]
+    output_name = [nd.name for nd in sess.get_outputs()]
+
+    def _get_feed_dict(feats_length):
+        return {'speech': np.zeros((1, feats_length, 560), dtype=np.float32), 'speech_lengths': np.array([feats_length,], dtype=np.int32)}
+
+    def _run(feed_dict):
+        output = sess.run(output_name, input_feed=feed_dict)
+        for name, value in zip(output_name, output):
+            print('{}: {}'.format(name, value.shape))
+
+    _run(_get_feed_dict(100))
+    _run(_get_feed_dict(200))

funasr_local/export/test/test_onnx_punc.py

@@ -0,0 +1,18 @@
+import onnxruntime
+import numpy as np
+
+
+if __name__ == '__main__':
+    onnx_path = "../damo/punc_ct-transformer_zh-cn-common-vocab272727-pytorch/model.onnx"
+    sess = onnxruntime.InferenceSession(onnx_path)
+    input_name = [nd.name for nd in sess.get_inputs()]
+    output_name = [nd.name for nd in sess.get_outputs()]
+
+    def _get_feed_dict(text_length):
+        return {'inputs': np.ones((1, text_length), dtype=np.int64), 'text_lengths': np.array([text_length,], dtype=np.int32)}
+
+    def _run(feed_dict):
+        output = sess.run(output_name, input_feed=feed_dict)
+        for name, value in zip(output_name, output):
+            print('{}: {}'.format(name, value))
+    _run(_get_feed_dict(10))

funasr_local/export/test/test_onnx_punc_vadrealtime.py

@@ -0,0 +1,22 @@
+import onnxruntime
+import numpy as np
+
+
+if __name__ == '__main__':
+    onnx_path = "./export/damo/punc_ct-transformer_zh-cn-common-vad_realtime-vocab272727/model.onnx"
+    sess = onnxruntime.InferenceSession(onnx_path)
+    input_name = [nd.name for nd in sess.get_inputs()]
+    output_name = [nd.name for nd in sess.get_outputs()]
+
+    def _get_feed_dict(text_length):
+        return {'inputs': np.ones((1, text_length), dtype=np.int64),
+                'text_lengths': np.array([text_length,], dtype=np.int32),
+                'vad_masks': np.ones((1, 1, text_length, text_length), dtype=np.float32),
+                'sub_masks': np.tril(np.ones((text_length, text_length), dtype=np.float32))[None, None, :, :].astype(np.float32)
+                }
+
+    def _run(feed_dict):
+        output = sess.run(output_name, input_feed=feed_dict)
+        for name, value in zip(output_name, output):
+            print('{}: {}'.format(name, value))
+    _run(_get_feed_dict(10))

funasr_local/export/test/test_onnx_vad.py

@@ -0,0 +1,26 @@
+import onnxruntime
+import numpy as np
+
+
+if __name__ == '__main__':
+    onnx_path = "/mnt/workspace/export/damo/speech_fsmn_vad_zh-cn-16k-common-pytorch/model.onnx"
+    sess = onnxruntime.InferenceSession(onnx_path)
+    input_name = [nd.name for nd in sess.get_inputs()]
+    output_name = [nd.name for nd in sess.get_outputs()]
+
+    def _get_feed_dict(feats_length):
+        
+        return {'speech': np.random.rand(1, feats_length, 400).astype(np.float32),
+                'in_cache0': np.random.rand(1, 128, 19, 1).astype(np.float32),
+                'in_cache1': np.random.rand(1, 128, 19, 1).astype(np.float32),
+                'in_cache2': np.random.rand(1, 128, 19, 1).astype(np.float32),
+                'in_cache3': np.random.rand(1, 128, 19, 1).astype(np.float32),
+                }
+
+    def _run(feed_dict):
+        output = sess.run(output_name, input_feed=feed_dict)
+        for name, value in zip(output_name, output):
+            print('{}: {}'.format(name, value.shape))
+
+    _run(_get_feed_dict(100))
+    _run(_get_feed_dict(200))

funasr_local/export/test/test_torchscripts.py

@@ -0,0 +1,17 @@
+import torch
+import numpy as np
+
+if __name__ == '__main__':
+	onnx_path = "/nfs/zhifu.gzf/export/damo/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-pytorch/model.torchscripts"
+	loaded = torch.jit.load(onnx_path)
+	
+	x = torch.rand([2, 21, 560])
+	x_len = torch.IntTensor([6, 21])
+	res = loaded(x, x_len)
+	print(res[0].size(), res[1])
+	
+	x = torch.rand([5, 50, 560])
+	x_len = torch.IntTensor([6, 21, 10, 30, 50])
+	res = loaded(x, x_len)
+	print(res[0].size(), res[1])
+