up

ami.py

@@ -209,6 +209,8 @@ _TRAIN_SAMPLE_IDS = [
     "TS3012d",
 ]
 
+_TRAIN_SAMPLES_BROKEN_SDM = ["IS1003b", "IS1007d"]
+
 _VALIDATION_SAMPLE_IDS = [
     "ES2011a",
     "ES2011c",
@@ -364,7 +366,10 @@ class AMI(datasets.GeneratorBasedBuilder):
                 text = " ".join(line_items[1:])
                 _, meeting_id, microphone_id, speaker_id, begin_time, end_time = _id.split("_")
 
-                if self.config.name == "sdm":
+                # sdm specific
+                if self.config.name == "sdm" and meeting_id in _TRAIN_SAMPLES_BROKEN_SDM:
+                    continue
+                elif self.config.name == "sdm":
                     microphone_id = "SDM1"
 
                 audio_filename = "_".join([split, _id.lower()]) + ".wav"