update

.gitignore

@@ -11,6 +11,7 @@
 /data/
 /docs/
 /dotenv/
+/examples/**/*.wav
 /trained_models/
 /temp/
 

examples/sample_filter/bad_case_find.py

@@ -51,7 +51,7 @@ def main():
         label1, prob1 = client.predict(
             audio=handle_file(filename),
             # model_name="vm_sound_classification8-ch32",
-            model_name="voicemail-id-id-2-ch4",
+            model_name="voicemail-en-ph-2-ch4",
             ground_true="Hello!!",
             api_name="/click_button"
         )
@@ -66,7 +66,7 @@ def main():
         )
         prob2 = float(prob2)
 
-        if label1 == "voicemail" and label2 in ("voicemail", "bell") and prob1 > 1.0:
+        if label1 == "voicemail" and label2 in ("voicemail", "bell") and prob1 > 0.6:
             pass
         elif label1 == "non_voicemail" and label2 not in ("voicemail", "bell") and prob1 > 0.6:
             pass

examples/sample_filter/correction.py

@@ -0,0 +1,70 @@
+#!/usr/bin/python3
+# -*- coding: utf-8 -*-
+import argparse
+from pathlib import Path
+import shutil
+
+from gradio_client import Client, handle_file
+from tqdm import tqdm
+
+from project_settings import project_path
+
+
+def get_args():
+    parser = argparse.ArgumentParser()
+    parser.add_argument(
+        "--data_dir",
+        default=r"E:\Users\tianx\HuggingDatasets\vm_sound_classification\data\wav_finished\en-PH\wav_finished",
+        type=str
+    )
+    parser.add_argument(
+        "--correction_dir",
+        default=r"E:\Users\tianx\HuggingDatasets\vm_sound_classification\data\correction",
+        type=str
+    )
+    args = parser.parse_args()
+    return args
+
+
+def main():
+    args = get_args()
+
+    data_dir = Path(args.data_dir)
+    correction_dir = Path(args.correction_dir)
+    correction_dir.mkdir(parents=True, exist_ok=True)
+
+    client = Client("http://127.0.0.1:7864/")
+
+    for idx, filename in tqdm(enumerate(data_dir.glob("**/*.wav"))):
+        # if idx < 200:
+        #     continue
+        ground_truth = filename.parts[-2]
+        filename = filename.as_posix()
+
+        label, prob = client.predict(
+            audio=handle_file(filename),
+            model_name="voicemail-en-ph-2-ch32",
+            ground_true="Hello!!",
+            api_name="/click_button"
+        )
+        prob = float(prob)
+
+        if label == "voicemail" and ground_truth in ("voicemail", "bell"):
+            pass
+        elif label == "non_voicemail" and ground_truth not in ("voicemail", "bell"):
+            pass
+        else:
+            print(f"ground_truth: {ground_truth}, label: {label}, prob: {prob}")
+
+            tgt_dir = correction_dir / ground_truth
+            tgt_dir.mkdir(parents=True, exist_ok=True)
+            shutil.move(
+                filename,
+                tgt_dir.as_posix(),
+            )
+
+    return
+
+
+if __name__ == '__main__':
+    main()

examples/sample_filter/wav_find_by_task_excel.py

@@ -13,12 +13,13 @@ from project_settings import project_path
 
 
 task_file_str = """
-task_DcTask_1_PH_LIVE_20250304_20250304-1.xlsx
-task_DcTask_1_PH_LIVE_20250305_20250305-1.xlsx
-task_DcTask_3_PH_LIVE_20250304_20250304-1.xlsx
-task_DcTask_3_PH_LIVE_20250305_20250305-1.xlsx
-task_DcTask_9_PH_LIVE_20250304_20250304-1.xlsx
-task_DcTask_9_PH_LIVE_20250305_20250305-1.xlsx
+task_DcTask_1_PH_LIVE_20250328_20250328-1.xlsx
+task_DcTask_1_PH_LIVE_20250329_20250329-1.xlsx
+task_DcTask_1_PH_LIVE_20250331_20250331-1.xlsx
+task_DcTask_3_PH_LIVE_20250328_20250328-1.xlsx
+task_DcTask_3_PH_LIVE_20250331_20250331-1.xlsx
+task_DcTask_9_PH_LIVE_20250329_20250329-1.xlsx
+task_DcTask_9_PH_LIVE_20250331_20250331-1.xlsx
 """
 
 

requirements.txt

@@ -9,5 +9,5 @@ tqdm==4.66.4
 overrides==1.9.0
 pyyaml==6.0.1
 evaluate==0.4.2
-gradio==4.44.1
+gradio
 python-dotenv==1.0.1