fix frame display style

app.py

@@ -105,6 +105,14 @@ def load(audio_path: str, converto="mel"):
     return list(norm(Xtr_spec))
 
 
+def format_second(seconds):
+    integer_part = int(seconds)
+    decimal_part = round(seconds - integer_part, 3)
+    hours, remainder = divmod(integer_part, 3600)
+    minutes, seconds = divmod(remainder, 60)
+    return f"{hours:02}:{minutes:02}:{seconds:02}.{decimal_part:.3f}"
+
+
 def infer(audio_path: str, log_name: str):
     if not audio_path:
         return None, "Please input an audio!"
@@ -113,6 +121,8 @@ def infer(audio_path: str, log_name: str):
     spec = log_name.split("_")[-1]
     try:
         input = load(audio_path, converto=spec)
+        dur = librosa.get_duration(path=audio_path)
+        frames_per_3s = input[0].shape[1]
         if "vit" in backbone or "swin" in backbone:
             eval_net = t_EvalNet(
                 backbone,
@@ -140,9 +150,14 @@ def infer(audio_path: str, log_name: str):
     for y in output:
         preds = list(y.T)
         for pred in preds:
+            start = index * TIME_LENGTH / frames_per_3s
+            if start > dur:
+                break
+
+            to = (index + 1) * TIME_LENGTH / frames_per_3s
             outputs.append(
                 {
-                    "Frame": index,
+                    "Frame": f"{format_second(start)} - {format_second(to)}",
                     "Tech": TRANSLATE[CLASSES[torch.argmax(pred).item()]],
                 }
             )