cleaning

lium_baseline/system.py

@@ -84,7 +84,7 @@ def init_clustering(init_diar, cep, model_cfg, vad_type="none"):
     """
 
     # Bic_lin is only applied when starting from a VAD segmentation or from UEM
-    if model_cfg["model"]["vad"]["type"] != "reference":
+    if vad_type != "reference":
         output_diar = bic_linear_segmentation(init_diar, cep, model_cfg)
     else:
         output_diar = init_diar
@@ -94,7 +94,7 @@ def init_clustering(init_diar, cep, model_cfg, vad_type="none"):
     output_diar = cluster.perform()
 
     # Viterbi devoding is only applied when starting from a VAD segmentation or from scratch
-    if model_cfg["model"]["vad"]["type"] != "reference":
+    if vad_type != "reference":
         output_diar = s4d.viterbi.viterbi_decoding(cep, output_diar, model_cfg['first_seg']['thr_vit'])
 
     return output_diar
@@ -135,14 +135,15 @@ def vec2link_xv(model_cfg, xv_vec, current_diar):
                                                check_missing=False)
 
     # Use 2 gaussian to shift the scores
-    if scores.modelset.shape[0] > 2:
-        th_w = customize_threshold(scores, th_w)
+    #if scores.modelset.shape[0] > 2:
+    #    th_w = customize_threshold(scores, th_w)
 
     scores.scoremat = 0.5 * (scores.scoremat + scores.scoremat.transpose())
 
     # Make the cluster names consistent
     for idx in range(len(scores.modelset)):
-        scores.modelset[idx] = current_diar[idx]["cluster"]
+        #scores.modelset[idx] = current_diar[idx]["cluster"]
+        scores.modelset[idx] = xv_vec.modelset[idx]
 
     ######################################################################################
     # MODIFIED AS WE NOW USE COSINE SIMILARITIES
@@ -305,8 +306,8 @@ def perform_second_seg(model,
         scores.scoremat = 0.5 * (scores.scoremat + scores.scoremat.transpose())
 
         # Calibration
-        if scores.modelset.shape[0] > 2:
-            th_w = customize_threshold(scores, th_w)
+        #if scores.modelset.shape[0] > 2:
+        #    th_w = customize_threshold(scores, th_w)
 
     # Run HAC clustering
     print(f"Avant HAC : len(diar)= {len(initial_diar.unique('cluster'))}, min et max scores: {scores.scoremat.min()} et {scores.scoremat.max()}, th = {th_w}")
@@ -794,7 +795,6 @@ def allies_init_seg(model, model_cfg, show, data_folder, verbose=False):
 
             init_diar.pad(15)
             init_diar.pack(25)
-            init_diar.pack(25)
 
         # Run the first pass of segmentation
         logger.info("\t* run 1st clustering")