init

attach_speaker_embedding_s2s.py

@@ -15,6 +15,7 @@ dataset = load_dataset(f"{hf_org}/{hf_dataset}", f"subset_{dataset_id}", split="
 audio_loader = Audio()
 se_model = os.getenv("SE_MODEL", "metavoice")
 max_seq_length = 10000000
+min_seq_length = 50000
 
 if se_model == "metavoice":
     from speaker_embedding_metavoice import MetaVoiceSE
@@ -39,7 +40,7 @@ def error_file(example):
     for side in sides:
         try:
             wav = audio_loader.decode_example(example[f"{side}.audio"])
-            if len(wav["array"]) == 0 or len(wav["array"]) > max_seq_length:
+            if len(wav["array"]) < min_seq_length or len(wav["array"]) > max_seq_length:
                 return False
         except ValueError:
             return False

main_s2s.sh

@@ -410,7 +410,7 @@ export DIRECTION="enA-zhA"
 python attach_speaker_embedding_s2s.py
 
 export SE_MODEL="w2vbert-600m"  # running
-export DATASET_ID=165
+export DATASET_ID=167
 export DIRECTION="enA-zhA"
 python attach_speaker_embedding_s2s.py