init

attach_speaker_embedding_s2s.py

@@ -14,7 +14,7 @@ hf_dataset = f"seamless-align-{direction}"
 dataset = load_dataset(f"{hf_org}/{hf_dataset}", f"subset_{dataset_id}", split="train")
 audio_loader = Audio()
 se_model = os.getenv("SE_MODEL", "metavoice")
-max_seq_length = 10000000
+max_seq_length = 1000000
 min_seq_length = 50000
 
 if se_model == "metavoice":
@@ -60,7 +60,7 @@ print(f"Num examples (after filtering): {len(dataset)}")
 
 def speaker_embedding(example):
     for side in sides:
-        print(len(example[f"{side}.audio"]["array"]))
+        # print(len(example[f"{side}.audio"]["array"]))
         embedding = speaker_embedder.get_speaker_embedding(
             example[f"{side}.audio"]["array"], example[f"{side}.audio"]["sampling_rate"]
         )