Upload diarization_pyannote_gilbert.py with huggingface_hub

diarization_pyannote_gilbert.py

@@ -287,13 +287,17 @@ def run_gilbert_diarization(
     # Convertir en segments
     segments = annotation_to_segments(annotation)
     
-    # Post-traitement Gilbert
-    print("[Gilbert] Post-traitement intelligent des segments...")
-    segments_processed = post_process_segments_gilbert(
-        segments,
-        min_segment_duration=min_segment_duration,
-        max_gap=merge_gaps
-    )
+    # Post-traitement Gilbert (optionnel - peut être désactivé)
+    if min_segment_duration > 0 or merge_gaps > 0:
+        print("[Gilbert] Post-traitement intelligent des segments...")
+        segments_processed = post_process_segments_gilbert(
+            segments,
+            min_segment_duration=min_segment_duration,
+            max_gap=merge_gaps
+        )
+    else:
+        print("[Gilbert] Post-traitement désactivé (segments bruts conservés)")
+        segments_processed = segments
     
     # Détecter les overlaps
     overlaps = detect_overlaps_gilbert(segments_processed)
@@ -401,8 +405,8 @@ def main():
     parser.add_argument("--max_speakers", type=int, default=None, help="Nombre maximum de locuteurs")
     parser.add_argument("--exclusive", action="store_true", help="Utiliser exclusive_speaker_diarization")
     parser.add_argument("--no-progress", action="store_true", help="Ne pas afficher la progression")
-    parser.add_argument("--min-segment", type=float, default=0.5, help="Durée minimale des segments (s)")
-    parser.add_argument("--merge-gaps", type=float, default=0.3, help="Gaps à fusionner (s)")
+    parser.add_argument("--min-segment", type=float, default=0.0, help="Durée minimale des segments (s). 0 = désactivé (recommandé pour meilleure précision)")
+    parser.add_argument("--merge-gaps", type=float, default=0.0, help="Gaps à fusionner (s). 0 = désactivé (recommandé pour meilleure précision)")
     
     args = parser.parse_args()