init

download_s2t_metadata.py

@@ -19,7 +19,7 @@ filename = os.path.basename(url)
 subprocess.run(["wget", url, "-O", filename])
 df = pd.read_csv(filename, sep='\t', header=None, dtype=str)
 df.columns = ["cc_warc", "cc_sha", "cc_document_url", "cc_lineno", "paragraph_digest", "sentence_digest", "text_lid_score", "laser_score", "direction", "side", "line_no"]
-df = df[df.side == "jpn"]
+df = df[df.side == direction_text]
 df["cc_lineno"] = df["cc_lineno"].astype(int)
 df.sort_values(by=["cc_warc", "cc_sha", "cc_document_url", "cc_lineno"], inplace=True)
 batch_size = int(len(df)/chunk_size)

main_s2s.sh

@@ -1,234 +1,387 @@
 ####################
-# enA-jaA: 718_606 #
+# enA-koA: 511_358 #
 ####################
 # test
 export DATASET_ID=test
-export DIRECTION="enA-jaA"
+export DIRECTION="enA-koA"
 export LINE_NO_START=0
 export LINE_NO_END=10
 python fetch_dataset_s2s.py
 # main
-for i in $(seq 1 144);
+for i in $(seq 1 50);
 do
   export N_POOL=15
   export DATASET_ID=${i}
-  export DIRECTION="enA-jaA"
+  export DIRECTION="enA-koA"
+  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 $(seq 51 100);
+do
+  export N_POOL=15
+  export DATASET_ID=${i}
+  export DIRECTION="enA-koA"
   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 1 144);
+for i in $(seq 1 20);
 do
+  export N_POOL=15
   export DATASET_ID=${i}
-  export DIRECTION="enA-jaA"
+  export DIRECTION="enA-koA"
+  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 $(seq 51 64);
+do
+  export N_POOL=15
+  export DATASET_ID=${i}
+  export DIRECTION="enA-koA"
+  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 1 144);
+for i in $(seq 1 20);
 do
   export DATASET_ID=${i}
-  export DIRECTION="enA-jaA"
+  export DIRECTION="enA-koA"
+  python attach_speaker_embedding_s2s.py
+done
+for i in $(seq 51 64);
+do
+  export DATASET_ID=${i}
+  export DIRECTION="enA-koA"
   python attach_speaker_embedding_s2s.py
 done
 
-####################
-# enA-hiA: 454_942 #
-####################
+######################
+# enA-zhA: 1_289_192 #
+######################
 # test
 export DATASET_ID=test
-export DIRECTION="enA-hiA"
+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 91);
+for i in $(seq 1 258);
 do
   export N_POOL=15
   export DATASET_ID=${i}
-  export DIRECTION="enA-hiA"
+  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 91);
+for i in $(seq 81 100);
 do
+  export N_POOL=15
   export DATASET_ID=${i}
-  export DIRECTION="enA-hiA"
+  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 91);
+for i in $(seq 81 100);
 do
   export DATASET_ID=${i}
-  export DIRECTION="enA-hiA"
+  export DIRECTION="enA-zhA"
   python attach_speaker_embedding_s2s.py
 done
-
-####################
-# enA-viA: 740_598 #
-####################
+######################
+# enA-frA: 3_054_258 #
+######################
 # test
 export DATASET_ID=test
-export DIRECTION="enA-viA"
+export DIRECTION="enA-frA"
 export LINE_NO_START=0
 export LINE_NO_END=10
 python fetch_dataset_s2s.py
 # main
-for i in $(seq 1 149);
+for i in $(seq 1 300);
 do
   export N_POOL=15
   export DATASET_ID=${i}
-  export DIRECTION="enA-viA"
+  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
+for i in $(seq 300 600);
+do
+  export N_POOL=15
+  export DATASET_ID=${i}
+  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 141 149);
+for i in $(seq 1 30);
 do
+  export N_POOL=15
   export DATASET_ID=${i}
-  export DIRECTION="enA-viA"
+  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
-for i in 25 26 33 34 35 36 37 38 39 40;
+for i in $(seq 300 330);
 do
+  export N_POOL=15
   export DATASET_ID=${i}
-  export DIRECTION="enA-viA"
+  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 1 144);
+for i in $(seq 1 30);
 do
   export DATASET_ID=${i}
-  export DIRECTION="enA-viA"
+  export DIRECTION="enA-frA"
   python attach_speaker_embedding_s2s.py
 done
-
-####################
-# enA-koA: 511_358 #
-####################
+# speaker embedding
+for i in $(seq 300 330);
+do
+  export DATASET_ID=${i}
+  export DIRECTION="enA-frA"
+  python attach_speaker_embedding_s2s.py
+done
+######################
+# enA-deA: 1_965_186 #
+######################
 # test
 export DATASET_ID=test
-export DIRECTION="enA-koA"
+export DIRECTION="deA-enA"
 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 $(seq 1 200);
 do
   export N_POOL=15
   export DATASET_ID=${i}
-  export DIRECTION="enA-koA"
+  export DIRECTION="deA-enA"
   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 51 100);
+for i in $(seq 201 394);
 do
   export N_POOL=15
   export DATASET_ID=${i}
-  export DIRECTION="enA-koA"
+  export DIRECTION="deA-enA"
   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
-
-######################
-# 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);
+# tokenize
+for i in $(seq 1 20);
 do
   export N_POOL=15
   export DATASET_ID=${i}
-  export DIRECTION="enA-zhA"
+  export DIRECTION="deA-enA"
   export LINE_NO_START=$(((DATASET_ID-1) * 2500))
   export LINE_NO_END=$((DATASET_ID * 2500))
   echo ${LINE_NO_START}
-  python fetch_dataset_s2s.py
+  python tokenize_dataset_s2s.py
+done
+for i in $(seq 200 240);
+do
+  export N_POOL=15
+  export DATASET_ID=${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
+# speaker embedding
+for i in $(seq 1 20);
+do
+  export DATASET_ID=${i}
+  export DIRECTION="deA-enA"
+  python attach_speaker_embedding_s2s.py
+done
+# speaker embedding
+for i in $(seq 200 240);
+do
+  export DATASET_ID=${i}
+  export DIRECTION="deA-enA"
+  python attach_speaker_embedding_s2s.py
 done
-
 ######################
-# enA-frA: 3_054_258 #
+# enA-esA: 2_658_022 #
 ######################
 # test
 export DATASET_ID=test
-export DIRECTION="enA-frA"
+export DIRECTION="enA-esA"
 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 $(seq 1 250);
 do
   export N_POOL=15
   export DATASET_ID=${i}
-  export DIRECTION="enA-frA"
+  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 300 600);
+for i in $(seq 251 532);
 do
   export N_POOL=15
   export DATASET_ID=${i}
-  export DIRECTION="enA-frA"
+  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
 
-######################
-# enA-esA: 2_658_022 #
-######################
+# tokenize
+for i in $(seq 1 80);
+do
+  export N_POOL=15
+  export DATASET_ID=${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 tokenize_dataset_s2s.py
+done
+# speaker embedding
+for i in $(seq 1 80);
+do
+  export DATASET_ID=${i}
+  export DIRECTION="enA-esA"
+  python attach_speaker_embedding_s2s.py
+done
+
+
+####################
+# enA-jaA: 718_606 #
+####################
 # test
 export DATASET_ID=test
-export DIRECTION="enA-esA"
+export DIRECTION="enA-jaA"
 export LINE_NO_START=0
 export LINE_NO_END=10
 python fetch_dataset_s2s.py
+# main
+for i in $(seq 1 144);
+do
+  export N_POOL=15
+  export DATASET_ID=${i}
+  export DIRECTION="enA-jaA"
+  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 $(seq 1 144);
+do
+  export DATASET_ID=${i}
+  export DIRECTION="enA-jaA"
+  python tokenize_dataset_s2s.py
+done
+# speaker embedding
+for i in $(seq 1 144);
+do
+  export DATASET_ID=${i}
+  export DIRECTION="enA-jaA"
+  python attach_speaker_embedding_s2s.py
+done
 
-######################
-# enA-deA: 1_965_186 #
-######################
+####################
+# enA-hiA: 454_942 #
+####################
 # test
 export DATASET_ID=test
-export DIRECTION="deA-enA"
+export DIRECTION="enA-hiA"
 export LINE_NO_START=0
 export LINE_NO_END=10
 python fetch_dataset_s2s.py
 # main
-for i in $(seq 1 200);
+for i in $(seq 1 91);
 do
   export N_POOL=15
   export DATASET_ID=${i}
-  export DIRECTION="deA-enA"
+  export DIRECTION="enA-hiA"
   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 201 394);
+# tokenize
+for i in $(seq 1 91);
+do
+  export DATASET_ID=${i}
+  export DIRECTION="enA-hiA"
+  python tokenize_dataset_s2s.py
+done
+# speaker embedding
+for i in $(seq 1 91);
+do
+  export DATASET_ID=${i}
+  export DIRECTION="enA-hiA"
+  python attach_speaker_embedding_s2s.py
+done
+####################
+# enA-viA: 740_598 #
+####################
+# test
+export DATASET_ID=test
+export DIRECTION="enA-viA"
+export LINE_NO_START=0
+export LINE_NO_END=10
+python fetch_dataset_s2s.py
+# main
+for i in $(seq 1 149);
 do
   export N_POOL=15
   export DATASET_ID=${i}
-  export DIRECTION="deA-enA"
+  export DIRECTION="enA-viA"
   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 $(seq 1 149);
+do
+  export DATASET_ID=${i}
+  export DIRECTION="enA-viA"
+  python tokenize_dataset_s2s.py
+done
+# speaker embedding
+for i in $(seq 1 149);
+do
+  export DATASET_ID=${i}
+  export DIRECTION="enA-viA"
+  python attach_speaker_embedding_s2s.py
 done

main_s2t.sh

@@ -21,6 +21,22 @@ make -j4
 alias wet_lines="${PWD}/bin/wet_lines"
 cd ../../
 
+
+######################
+# enA-est: 734_146 #
+######################
+# text
+export DIRECTION_SPEECH="enA"
+export DIRECTION_TEXT="est"
+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-jpn: 734_146 #
 ######################
@@ -35,7 +51,7 @@ do
 done
 python format_text.py
 # audio
-for i in $(seq 61 120);
+for i in $(seq 140 294);
 do
   export N_POOL=15
   export DATASET_ID=${i}
@@ -46,6 +62,7 @@ do
   echo ${LINE_NO_START}
   python fetch_dataset_s2t.py
 done
+
 ########
 # NLLB #
 ########

speaker_embedding_clap.py

@@ -0,0 +1,35 @@
+"""CLAP embedding.
+- feature dimension: 512
+- source: https://huggingface.co/laion/larger_clap_music_and_speech
+"""
+from typing import Optional
+
+import torch
+import librosa
+import numpy as np
+from transformers import ClapModel, ClapProcessor
+
+
+class ClapSE:
+    def __init__(self, ckpt: str = "laion/larger_clap_music_and_speech"):
+        self.model = ClapModel.from_pretrained(ckpt)
+        self.device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+        self.model.to(self.device)
+        self.model.eval()
+        self.processor = ClapProcessor.from_pretrained(ckpt)
+
+    def get_speaker_embedding(self, wav: np.ndarray, sampling_rate: Optional[int] = None) -> np.ndarray:
+        if sampling_rate != self.processor.feature_extractor.sampling_rate:
+            wav = librosa.resample(wav, orig_sr=sampling_rate, target_sr=self.processor.feature_extractor.sampling_rate)
+        inputs = self.processor(
+            audios=wav, sampling_rate=self.processor.feature_extractor.sampling_rate, return_tensors="pt"
+        )
+        with torch.no_grad():
+            outputs = self.model.get_audio_features(**{k: v.to(self.device) for k, v in inputs.items()})
+        return outputs.cpu().numpy()[0]
+
+
+class ClapGeneralSE(ClapSE):
+
+    def __init__(self):
+        super().__init__(ckpt="laion/larger_clap_general")