init

attach_speaker_embedding_s2s.py

@@ -5,9 +5,6 @@ import shutil
 from soundfile import LibsndfileError
 from datasets import load_dataset, DatasetDict, Audio
 
-from speaker_embedding_metavoice import MetaVoiceSE
-
-
 direction = os.getenv("DIRECTION", "enA-jaA")
 sides = set(direction.split("-"))
 dataset_id = os.getenv("DATASET_ID", 0)
@@ -16,7 +13,15 @@ hf_org = os.getenv("HF_ORG", "asahi417")
 hf_dataset = f"seamless-align-{direction}"
 dataset = load_dataset(f"{hf_org}/{hf_dataset}", f"subset_{dataset_id}", split="train")
 audio_loader = Audio()
-speaker_embedder = MetaVoiceSE()
+se_model = os.getenv("SE_MODEL", "metavoice")
+if se_model == "metavoice":
+    from speaker_embedding_metavoice import MetaVoiceSE
+    speaker_embedder = MetaVoiceSE()
+elif se_model == "pyannote":
+    from speaker_embedding_pyannote import PyannoteSE
+    speaker_embedder = PyannoteSE()
+else:
+    raise ValueError(f"unknown speaker embedding: {se_model}")
 
 
 def error_file(example):
@@ -51,7 +56,7 @@ dataset = dataset.map(
     num_proc=num_proc,
     desc="attach speaker embedding dataset"
 )
-DatasetDict({"train": dataset}).push_to_hub(f"{hf_org}/{hf_dataset}.speaker-embedding.metavoice", config_name=f"subset_{dataset_id}")
+DatasetDict({"train": dataset}).push_to_hub(f"{hf_org}/{hf_dataset}.speaker-embedding.{se_model}", config_name=f"subset_{dataset_id}")
 cache_dir = f"{expanduser('~')}/.cache/huggingface/datasets/{hf_org}___{hf_dataset}/subset_{dataset_id}"
 if os.path.exists(cache_dir):
     shutil.rmtree(cache_dir)

main_s2s.sh

@@ -26,13 +26,13 @@ do
   python tokenize_dataset_s2s.py
 done
 # speaker embedding
+export SE_MODEL="metavoice"
 for i in $(seq 1 144);
 do
   export DATASET_ID=${i}
   export DIRECTION="enA-jaA"
   python attach_speaker_embedding_s2s.py
 done
-
 for i in $(seq 2 40);
 do
   export DATASET_ID=${i}
@@ -45,15 +45,12 @@ do
   export DIRECTION="enA-jaA"
   python attach_speaker_embedding_s2s.py
 done
-
 for i in $(seq 81 120);
 do
   export DATASET_ID=${i}
   export DIRECTION="enA-jaA"
   python attach_speaker_embedding_s2s.py
 done
-
-
 for i in $(seq 121 144);
 do
   export DATASET_ID=${i}
@@ -109,16 +106,6 @@ do
   echo ${LINE_NO_START}
   python fetch_dataset_s2s.py
 done
-for i in 114 77 78 79 80;
-do
-  export N_POOL=15
-  export DATASET_ID=${i}
-  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 120 140);
 do

speaker_embedding_pyannote.py

@@ -0,0 +1,36 @@
+"""Pyannote speaker embedding model.
+- pip install pyannote.audio
+- feature dimension: 512
+- source: https://huggingface.co/pyannote/embedding
+"""
+from typing import Optional, Union, Tuple
+import torch
+import numpy as np
+from pyannote.audio import Model
+from pyannote.audio import Inference
+from pyannote.audio.core.inference import fix_reproducibility, map_with_specifications
+
+
+class PyannoteSE:
+
+    def __init__(self):
+        model = Model.from_pretrained("pyannote/embedding")
+        self.device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+        model.to(self.device)
+        model.eval()
+        self.inference = Inference(model, window="whole")
+
+    def get_speaker_embedding(self, wav: np.ndarray, sampling_rate: Optional[int] = None) -> np.ndarray:
+        wav = torch.as_tensor(wav.reshape(1, -1), dtype=torch.float32).to(self.device)
+        fix_reproducibility(self.inference.device)
+        if self.inference.window == "sliding":
+            return self.inference.slide(wav, sampling_rate, hook=None)
+
+        outputs: Union[np.ndarray, Tuple[np.ndarray]] = self.inference.infer(wav[None])
+
+        def __first_sample(outputs: np.ndarray, **kwargs) -> np.ndarray:
+            return outputs[0]
+
+        return map_with_specifications(
+            self.inference.model.specifications, __first_sample, outputs
+        )

tokenize_dataset_s2s.py

@@ -54,6 +54,7 @@ dataset = dataset.map(
     desc="tokenize dataset"
 )
 DatasetDict({"train": dataset}).push_to_hub(f"{hf_org}/{hf_dataset}.tokenized", config_name=f"subset_{dataset_id}")
+# DatasetDict({"train": dataset}).push_to_hub(f"{hf_org}/{hf_dataset}.tokenized.encodec", config_name=f"subset_{dataset_id}")
 cache_dir = f"{expanduser('~')}/.cache/huggingface/datasets/{hf_org}___{hf_dataset}/subset_{dataset_id}"
 if os.path.exists(cache_dir):
     shutil.rmtree(cache_dir)