Fix a bug with speaker names in left sidebar

app.py

@@ -322,24 +322,37 @@ try:
         stem = fname.rsplit(".", 1)[0]   # strip extension
         return (stem[:5] + "...") if len(stem) > 5 else stem
 
-    token_display_map = {}
+    # Build token_display_map in two passes to guarantee globally unique display keys.
+    # Pass 1: collect all (display_label -> raw_token) pairs across every file.
+    # Pass 2: any display label that appears more than once (collision between
+    #         files that share the same short prefix AND speaker name) gets the
+    #         raw speaker ID appended to disambiguate.
+    raw_pairs = []   # [(display_label, raw_token), ...]
     for fn in st.session_state.file_names:
         if not (fn in st.session_state.results and len(st.session_state.results[fn]) == 2):
             continue
         short = _short_fn(fn)
-        # First pass: collect display labels per file to detect duplicates
-        sp_labels = {sp: f"{short}: {get_display_name(sp, fn)}"
-                     for sp in st.session_state.results[fn][0].labels()}
-        label_counts = {}
-        for disp in sp_labels.values():
-            label_counts[disp] = label_counts.get(disp, 0) + 1
-        # Second pass: build map, disambiguating where needed
-        for sp, disp in sp_labels.items():
-            raw = f"{fn}: {sp}"          # raw token always uses full filename
-            if label_counts[disp] > 1:
-                token_display_map[f"{short}: {get_display_name(sp, fn)} ({sp})"] = raw
-            else:
-                token_display_map[disp] = raw
+        for sp in st.session_state.results[fn][0].labels():
+            raw   = f"{fn}: {sp}"
+            disp  = f"{short}: {get_display_name(sp, fn)}"
+            raw_pairs.append((disp, raw))
+
+    # Count occurrences of each display label globally
+    disp_counts = {}
+    for disp, _ in raw_pairs:
+        disp_counts[disp] = disp_counts.get(disp, 0) + 1
+
+    token_display_map = {}
+    for disp, raw in raw_pairs:
+        fn, sp = raw.split(": ", 1)
+        short  = _short_fn(fn)
+        if disp_counts[disp] > 1:
+            # Disambiguate with raw speaker ID
+            unique_disp = f"{short}: {get_display_name(sp, fn)} ({sp})"
+        else:
+            unique_disp = disp
+        token_display_map[unique_disp] = raw
+
     display_speaker_tokens = list(token_display_map.keys())
 
     # -----------------------------------------------------------------------