update

examples/sound_classification_by_lstm/step_6_export_onnx_model.py

@@ -120,9 +120,9 @@ def main():
         "logits", "new_h", "new_c"
     ]
     logits, new_h, new_c = ort_session.run(output_names, input_feed)
-    print(f"logits: {logits.shape}")
-    print(f"new_h: {new_h.shape}")
-    print(f"new_c: {new_c.shape}")
+    # print(f"logits: {logits.shape}")
+    # print(f"new_h: {new_h.shape}")
+    # print(f"new_c: {new_c.shape}")
     return
 
 

examples/sound_classification_by_lstm/step_8_test_onnx_model.py

@@ -31,7 +31,8 @@ def get_args():
     )
     parser.add_argument(
         "--wav_file",
-        default=r"C:\Users\tianx\Desktop\a073d03d-d280-46df-9b2d-d904965f4500_zh-CN_h3f25ivhb0c0_1719478037746.wav",
+        # default=r"C:\Users\tianx\Desktop\a073d03d-d280-46df-9b2d-d904965f4500_zh-CN_h3f25ivhb0c0_1719478037746.wav",
+        default=(project_path / "data/examples/examples/zh-TW/voicemail/00a1d109-23c2-4b8b-a066-993ac2ae8260_zh-TW_1672210785598.wav").as_posix(),
         type=str
     )
 
@@ -107,10 +108,23 @@ def main():
         "logits", "new_h", "new_c"
     ]
     logits, new_h, new_c = ort_session.run(output_names, input_feed)
-    print(f"logits: {logits.shape}")
-    print(f"new_h: {new_h.shape}")
-    print(f"new_c: {new_c.shape}")
+    # print(f"logits: {logits.shape}")
+    # print(f"new_h: {new_h.shape}")
+    # print(f"new_c: {new_c.shape}")
 
+    logits = torch.tensor(logits, dtype=torch.float32)
+    probs = torch.nn.functional.softmax(logits, dim=-1)
+    label_idx = torch.argmax(probs, dim=-1)
+
+    label_idx = label_idx.cpu()
+    probs = probs.cpu()
+
+    label_idx = label_idx.numpy()[0]
+    prob = probs.numpy()[0][label_idx]
+
+    label_str = vocabulary.get_token_from_index(label_idx, namespace="labels")
+    print(label_str)
+    print(prob)
     return
 
 

tabs/cls_tab.py

@@ -1,11 +1,14 @@
 #!/usr/bin/python3
 # -*- coding: utf-8 -*-
 import argparse
+import json
 from functools import lru_cache
+from os import times
 from pathlib import Path
 import platform
 import shutil
 import tempfile
+import time
 import zipfile
 from typing import Tuple
 
@@ -61,10 +64,12 @@ def when_click_cls_button(audio_t,
     inputs = torch.tensor(inputs, dtype=torch.float32)
     inputs = torch.unsqueeze(inputs, dim=0)
 
+    time_begin = time.time()
     with torch.no_grad():
         logits = model.forward(inputs)
         probs = torch.nn.functional.softmax(logits, dim=-1)
         label_idx = torch.argmax(probs, dim=-1)
+    time_cost = time.time() - time_begin
 
     label_idx = label_idx.cpu()
     probs = probs.cpu()
@@ -74,7 +79,13 @@ def when_click_cls_button(audio_t,
 
     label_str = vocabulary.get_token_from_index(label_idx, namespace="labels")
 
-    return label_str, round(prob, 4)
+    result = {
+        "label": label_str,
+        "prob": round(float(prob), 4),
+        "time_cost": round(time_cost, 4),
+    }
+    result = json.dumps(result, ensure_ascii=False, indent=4)
+    return result
 
 
 def get_cls_tab(examples_dir: str, trained_model_dir: str):
@@ -121,13 +132,12 @@ def get_cls_tab(examples_dir: str, trained_model_dir: str):
 
                 cls_button = gr.Button("run", variant="primary")
             with gr.Column(scale=3):
-                cls_label = gr.Textbox(label="label")
-                cls_probability = gr.Number(label="probability")
+                cls_outputs = gr.Textbox(label="outputs", lines=1, max_lines=15)
 
         gr.Examples(
             cls_examples,
             inputs=[cls_audio, cls_model_name, cls_ground_true],
-            outputs=[cls_label, cls_probability],
+            outputs=[cls_outputs],
             fn=when_click_cls_button,
             examples_per_page=5,
         )
@@ -135,7 +145,7 @@ def get_cls_tab(examples_dir: str, trained_model_dir: str):
         cls_button.click(
             when_click_cls_button,
             inputs=[cls_audio, cls_model_name, cls_ground_true],
-            outputs=[cls_label, cls_probability],
+            outputs=[cls_outputs],
         )
 
     return locals()