init

attach_speaker_embedding_s2s.py

@@ -21,8 +21,8 @@ elif se_model == "pyannote":
     from speaker_embedding_pyannote import PyannoteSE
     speaker_embedder = PyannoteSE()
 elif se_model == "w2vbert-600m":
-    from speaker_embedding_hf import Wav2VecEmbedding
-    speaker_embedder = Wav2VecEmbedding()
+    from speaker_embedding_hf import W2VBERTEmbedding
+    speaker_embedder = W2VBERTEmbedding()
 elif se_model == "xlsr-2b":
     from speaker_embedding_hf import XLSR2BEmbedding
     speaker_embedder = XLSR2BEmbedding()

main_s2s.sh

@@ -337,6 +337,8 @@ do
   python tokenize_dataset_s2s.py
 done
 # speaker embedding
+export SE_MODEL="metavoice"
+export SE_MODEL="w2vbert-600m"
 for i in $(seq 1 258);
 do
   export DATASET_ID=${i}
@@ -371,6 +373,8 @@ do
   python tokenize_dataset_s2s.py
 done
 # speaker embedding
+export SE_MODEL="metavoice"
+export SE_MODEL="w2vbert-600m"
 for i in $(seq 1 144);
 do
   export DATASET_ID=${i}
@@ -406,6 +410,8 @@ do
   python tokenize_dataset_s2s.py
 done
 # speaker embedding
+export SE_MODEL="metavoice"
+export SE_MODEL="w2vbert-600m"
 for i in $(seq 1 91);
 do
   export DATASET_ID=${i}
@@ -440,6 +446,8 @@ do
   python tokenize_dataset_s2s.py
 done
 # speaker embedding
+export SE_MODEL="metavoice"
+export SE_MODEL="w2vbert-600m"
 for i in $(seq 1 149);
 do
   export DATASET_ID=${i}