init

main_s2s.sh

@@ -8,7 +8,7 @@ export LINE_NO_START=0
 export LINE_NO_END=10
 python fetch_dataset_s2s.py
 # main
-for i in $(seq 1 50);
+for i in $(seq 34 50);
 do
   export N_POOL=15
   export DATASET_ID=${i}
@@ -18,7 +18,7 @@ do
   echo ${LINE_NO_START}
   python fetch_dataset_s2s.py
 done
-for i in $(seq 51 100);
+for i in $(seq 73 100);
 do
   export N_POOL=15
   export DATASET_ID=${i}
@@ -29,7 +29,7 @@ do
   python fetch_dataset_s2s.py
 done
 # tokenize
-for i in $(seq 1 20);
+for i in $(seq 21 36);
 do
   export N_POOL=15
   export DATASET_ID=${i}
@@ -39,7 +39,7 @@ do
   echo ${LINE_NO_START}
   python tokenize_dataset_s2s.py
 done
-for i in $(seq 51 64);
+for i in $(seq 65 75);
 do
   export N_POOL=15
   export DATASET_ID=${i}
@@ -64,54 +64,36 @@ do
 done
 
 ######################
-# enA-zhA: 1_289_192 #
+# enA-frA: 3_054_258 #
 ######################
 # test
 export DATASET_ID=test
-export DIRECTION="enA-zhA"
+export DIRECTION="enA-frA"
 export LINE_NO_START=0
 export LINE_NO_END=10
 python fetch_dataset_s2s.py
 # main
-for i in $(seq 1 258);
+for i in $(seq 106 200);
 do
   export N_POOL=15
   export DATASET_ID=${i}
-  export DIRECTION="enA-zhA"
+  export DIRECTION="enA-frA"
   export LINE_NO_START=$(((DATASET_ID-1) * 2500))
   export LINE_NO_END=$((DATASET_ID * 2500))
   echo ${LINE_NO_START}
   python fetch_dataset_s2s.py
 done
-# tokenize
-for i in $(seq 101 150);
+for i in $(seq 201 300);
 do
   export N_POOL=15
   export DATASET_ID=${i}
-  export DIRECTION="enA-zhA"
+  export DIRECTION="enA-frA"
   export LINE_NO_START=$(((DATASET_ID-1) * 2500))
   export LINE_NO_END=$((DATASET_ID * 2500))
   echo ${LINE_NO_START}
-  python tokenize_dataset_s2s.py
-done
-# speaker embedding
-for i in $(seq 101 150);
-do
-  export DATASET_ID=${i}
-  export DIRECTION="enA-zhA"
-  python attach_speaker_embedding_s2s.py
+  python fetch_dataset_s2s.py
 done
-######################
-# enA-frA: 3_054_258 #
-######################
-# test
-export DATASET_ID=test
-export DIRECTION="enA-frA"
-export LINE_NO_START=0
-export LINE_NO_END=10
-python fetch_dataset_s2s.py
-# main
-for i in $(seq 1 300);
+for i in $(seq 427 500);
 do
   export N_POOL=15
   export DATASET_ID=${i}
@@ -121,7 +103,7 @@ do
   echo ${LINE_NO_START}
   python fetch_dataset_s2s.py
 done
-for i in $(seq 300 600);
+for i in $(seq 500 600);
 do
   export N_POOL=15
   export DATASET_ID=${i}
@@ -132,7 +114,7 @@ do
   python fetch_dataset_s2s.py
 done
 # tokenize
-for i in $(seq 31 60);
+for i in $(seq 61 105);
 do
   export N_POOL=15
   export DATASET_ID=${i}
@@ -142,7 +124,15 @@ do
   echo ${LINE_NO_START}
   python tokenize_dataset_s2s.py
 done
-for i in $(seq 331 360);
+# speaker embedding
+for i in $(seq 61 105);
+do
+  export DATASET_ID=${i}
+  export DIRECTION="enA-frA"
+  python attach_speaker_embedding_s2s.py
+done
+# tokenize
+for i in $(seq 361 425);
 do
   export N_POOL=15
   export DATASET_ID=${i}
@@ -153,19 +143,14 @@ do
   python tokenize_dataset_s2s.py
 done
 # speaker embedding
-for i in $(seq 31 60);
-do
-  export DATASET_ID=${i}
-  export DIRECTION="enA-frA"
-  python attach_speaker_embedding_s2s.py
-done
-# speaker embedding
-for i in $(seq 331 360);
+for i in $(seq 361 425);
 do
   export DATASET_ID=${i}
   export DIRECTION="enA-frA"
   python attach_speaker_embedding_s2s.py
 done
+
+
 ######################
 # enA-deA: 1_965_186 #
 ######################
@@ -197,18 +182,7 @@ do
   python fetch_dataset_s2s.py
 done
 # tokenize
-for i in 215 216 232;
-do
-  export N_POOL=15
-  export DATASET_ID=${i}
-  export DIRECTION="deA-enA"
-  export LINE_NO_START=$(((DATASET_ID-1) * 2500))
-  export LINE_NO_END=$((DATASET_ID * 2500))
-  echo ${LINE_NO_START}
-  python tokenize_dataset_s2s.py
-
-done
-for i in $(seq 21 50);
+for i in $(seq 51 104);
 do
   export N_POOL=15
   export DATASET_ID=${i}
@@ -218,7 +192,7 @@ do
   echo ${LINE_NO_START}
   python tokenize_dataset_s2s.py
 done
-for i in $(seq 241 290);
+for i in $(seq 370 394);
 do
   export N_POOL=15
   export DATASET_ID=${i}
@@ -229,14 +203,14 @@ do
   python tokenize_dataset_s2s.py
 done
 # speaker embedding
-for i in $(seq 21 50);
+for i in $(seq 28 104);
 do
   export DATASET_ID=${i}
   export DIRECTION="deA-enA"
   python attach_speaker_embedding_s2s.py
 done
 # speaker embedding
-for i in $(seq 241 290);
+for i in $(seq 370 394);
 do
   export DATASET_ID=${i}
   export DIRECTION="deA-enA"
@@ -252,7 +226,7 @@ export LINE_NO_START=0
 export LINE_NO_END=10
 python fetch_dataset_s2s.py
 # main
-for i in $(seq 1 250);
+for i in $(seq 54 125);
 do
   export N_POOL=15
   export DATASET_ID=${i}
@@ -262,7 +236,27 @@ do
   echo ${LINE_NO_START}
   python fetch_dataset_s2s.py
 done
-for i in $(seq 251 532);
+for i in $(seq 126 250);
+do
+  export N_POOL=15
+  export DATASET_ID=${i}
+  export DIRECTION="enA-esA"
+  export LINE_NO_START=$(((DATASET_ID-1) * 2500))
+  export LINE_NO_END=$((DATASET_ID * 2500))
+  echo ${LINE_NO_START}
+  python fetch_dataset_s2s.py
+done
+for i in $(seq 320 420);
+do
+  export N_POOL=15
+  export DATASET_ID=${i}
+  export DIRECTION="enA-esA"
+  export LINE_NO_START=$(((DATASET_ID-1) * 2500))
+  export LINE_NO_END=$((DATASET_ID * 2500))
+  echo ${LINE_NO_START}
+  python fetch_dataset_s2s.py
+done
+for i in $(seq 421 532);
 do
   export N_POOL=15
   export DATASET_ID=${i}
@@ -272,7 +266,6 @@ do
   echo ${LINE_NO_START}
   python fetch_dataset_s2s.py
 done
-
 # tokenize
 for i in $(seq 1 50);
 do
@@ -291,7 +284,6 @@ do
   export DIRECTION="enA-esA"
   python attach_speaker_embedding_s2s.py
 done
-
 # tokenize
 for i in $(seq 251 280);
 do
@@ -312,6 +304,45 @@ do
 done
 
 
+
+######################
+# enA-zhA: 1_289_192 #
+######################
+# test
+export DATASET_ID=test
+export DIRECTION="enA-zhA"
+export LINE_NO_START=0
+export LINE_NO_END=10
+python fetch_dataset_s2s.py
+# main
+for i in $(seq 1 258);
+do
+  export N_POOL=15
+  export DATASET_ID=${i}
+  export DIRECTION="enA-zhA"
+  export LINE_NO_START=$(((DATASET_ID-1) * 2500))
+  export LINE_NO_END=$((DATASET_ID * 2500))
+  echo ${LINE_NO_START}
+  python fetch_dataset_s2s.py
+done
+# tokenize
+for i in $(seq 1 258);
+do
+  export N_POOL=15
+  export DATASET_ID=${i}
+  export DIRECTION="enA-zhA"
+  export LINE_NO_START=$(((DATASET_ID-1) * 2500))
+  export LINE_NO_END=$((DATASET_ID * 2500))
+  echo ${LINE_NO_START}
+  python tokenize_dataset_s2s.py
+done
+# speaker embedding
+for i in $(seq 1 258);
+do
+  export DATASET_ID=${i}
+  export DIRECTION="enA-zhA"
+  python attach_speaker_embedding_s2s.py
+done
 ####################
 # enA-jaA: 718_606 #
 ####################

main_s2t.sh

@@ -22,9 +22,23 @@ alias wet_lines="${PWD}/bin/wet_lines"
 cd ../../
 
 
-######################
-# enA-est: 734_146 #
-######################
+###########
+# enA-vie #
+###########
+# text
+export DIRECTION_SPEECH="enA"
+export DIRECTION_TEXT="vie"
+export CHUNK_SIZE=20
+python download_s2t_metadata.py
+for i in $(seq 1 ${CHUNK_SIZE});
+do
+  cat seamless.dataset.metadata.public.${DIRECTION_SPEECH}-${DIRECTION_TEXT}.withduration.reordered.batch_${i}.tsv | egrep ^crawl-data | tr '\t' ' ' | wet_lines | tee metadata.${DIRECTION_SPEECH}-${DIRECTION_TEXT}.batch_${i}.tsv &
+done
+python format_text.py
+
+###########
+# enA-est #
+###########
 # text
 export DIRECTION_SPEECH="enA"
 export DIRECTION_TEXT="est"
@@ -36,7 +50,7 @@ do
 done
 python format_text.py
 # audio
-for i in $(seq 1 300);
+for i in $(seq 19 300);
 do
   export N_POOL=15
   export DATASET_ID=${i}
@@ -49,9 +63,9 @@ do
 done
 
 
-######################
-# enA-jpn: 734_146 #
-######################
+###########
+# enA-jpn #
+###########
 # text
 export DIRECTION_SPEECH="enA"
 export DIRECTION_TEXT="jpn"
@@ -63,7 +77,7 @@ do
 done
 python format_text.py
 # audio
-for i in $(seq 1 294);
+for i in $(seq 233 294);
 do
   export N_POOL=15
   export DATASET_ID=${i}

tokenizer_encodec.py

@@ -87,6 +87,7 @@ class BaseEncodecTokenizer:
         # tokenize by chunk in a sequential manner
         for offset in tqdm(list(range(0, input_length - chunk_length + stride, stride))):
             frame = wav[:, :1, offset: offset + chunk_length]
+            print(frame.shape)
             tmp_tokens, _ = self.codec_model.encode(frame.to(self.device))
             offset_latent = int(offset / self.sample_per_token)
             tmp_tokens = tmp_tokens.to("cpu") if cpu_offload else tmp_tokens.to(original_device)