asahi417 commited on
Commit
bb3ef72
1 Parent(s): 96a44aa
attach_speaker_embedding_s2s.py CHANGED
@@ -5,9 +5,6 @@ import shutil
5
  from soundfile import LibsndfileError
6
  from datasets import load_dataset, DatasetDict, Audio
7
 
8
- from speaker_embedding_metavoice import MetaVoiceSE
9
-
10
-
11
  direction = os.getenv("DIRECTION", "enA-jaA")
12
  sides = set(direction.split("-"))
13
  dataset_id = os.getenv("DATASET_ID", 0)
@@ -16,7 +13,15 @@ hf_org = os.getenv("HF_ORG", "asahi417")
16
  hf_dataset = f"seamless-align-{direction}"
17
  dataset = load_dataset(f"{hf_org}/{hf_dataset}", f"subset_{dataset_id}", split="train")
18
  audio_loader = Audio()
19
- speaker_embedder = MetaVoiceSE()
 
 
 
 
 
 
 
 
20
 
21
 
22
  def error_file(example):
@@ -51,7 +56,7 @@ dataset = dataset.map(
51
  num_proc=num_proc,
52
  desc="attach speaker embedding dataset"
53
  )
54
- DatasetDict({"train": dataset}).push_to_hub(f"{hf_org}/{hf_dataset}.speaker-embedding.metavoice", config_name=f"subset_{dataset_id}")
55
  cache_dir = f"{expanduser('~')}/.cache/huggingface/datasets/{hf_org}___{hf_dataset}/subset_{dataset_id}"
56
  if os.path.exists(cache_dir):
57
  shutil.rmtree(cache_dir)
 
5
  from soundfile import LibsndfileError
6
  from datasets import load_dataset, DatasetDict, Audio
7
 
 
 
 
8
  direction = os.getenv("DIRECTION", "enA-jaA")
9
  sides = set(direction.split("-"))
10
  dataset_id = os.getenv("DATASET_ID", 0)
 
13
  hf_dataset = f"seamless-align-{direction}"
14
  dataset = load_dataset(f"{hf_org}/{hf_dataset}", f"subset_{dataset_id}", split="train")
15
  audio_loader = Audio()
16
+ se_model = os.getenv("SE_MODEL", "metavoice")
17
+ if se_model == "metavoice":
18
+ from speaker_embedding_metavoice import MetaVoiceSE
19
+ speaker_embedder = MetaVoiceSE()
20
+ elif se_model == "pyannote":
21
+ from speaker_embedding_pyannote import PyannoteSE
22
+ speaker_embedder = PyannoteSE()
23
+ else:
24
+ raise ValueError(f"unknown speaker embedding: {se_model}")
25
 
26
 
27
  def error_file(example):
 
56
  num_proc=num_proc,
57
  desc="attach speaker embedding dataset"
58
  )
59
+ DatasetDict({"train": dataset}).push_to_hub(f"{hf_org}/{hf_dataset}.speaker-embedding.{se_model}", config_name=f"subset_{dataset_id}")
60
  cache_dir = f"{expanduser('~')}/.cache/huggingface/datasets/{hf_org}___{hf_dataset}/subset_{dataset_id}"
61
  if os.path.exists(cache_dir):
62
  shutil.rmtree(cache_dir)
main_s2s.sh CHANGED
@@ -26,13 +26,13 @@ do
26
  python tokenize_dataset_s2s.py
27
  done
28
  # speaker embedding
 
29
  for i in $(seq 1 144);
30
  do
31
  export DATASET_ID=${i}
32
  export DIRECTION="enA-jaA"
33
  python attach_speaker_embedding_s2s.py
34
  done
35
-
36
  for i in $(seq 2 40);
37
  do
38
  export DATASET_ID=${i}
@@ -45,15 +45,12 @@ do
45
  export DIRECTION="enA-jaA"
46
  python attach_speaker_embedding_s2s.py
47
  done
48
-
49
  for i in $(seq 81 120);
50
  do
51
  export DATASET_ID=${i}
52
  export DIRECTION="enA-jaA"
53
  python attach_speaker_embedding_s2s.py
54
  done
55
-
56
-
57
  for i in $(seq 121 144);
58
  do
59
  export DATASET_ID=${i}
@@ -109,16 +106,6 @@ do
109
  echo ${LINE_NO_START}
110
  python fetch_dataset_s2s.py
111
  done
112
- for i in 114 77 78 79 80;
113
- do
114
- export N_POOL=15
115
- export DATASET_ID=${i}
116
- export DIRECTION="enA-viA"
117
- export LINE_NO_START=$(((DATASET_ID-1) * 2500))
118
- export LINE_NO_END=$((DATASET_ID * 2500))
119
- echo ${LINE_NO_START}
120
- python fetch_dataset_s2s.py
121
- done
122
  # tokenize
123
  for i in $(seq 120 140);
124
  do
 
26
  python tokenize_dataset_s2s.py
27
  done
28
  # speaker embedding
29
+ export SE_MODEL="metavoice"
30
  for i in $(seq 1 144);
31
  do
32
  export DATASET_ID=${i}
33
  export DIRECTION="enA-jaA"
34
  python attach_speaker_embedding_s2s.py
35
  done
 
36
  for i in $(seq 2 40);
37
  do
38
  export DATASET_ID=${i}
 
45
  export DIRECTION="enA-jaA"
46
  python attach_speaker_embedding_s2s.py
47
  done
 
48
  for i in $(seq 81 120);
49
  do
50
  export DATASET_ID=${i}
51
  export DIRECTION="enA-jaA"
52
  python attach_speaker_embedding_s2s.py
53
  done
 
 
54
  for i in $(seq 121 144);
55
  do
56
  export DATASET_ID=${i}
 
106
  echo ${LINE_NO_START}
107
  python fetch_dataset_s2s.py
108
  done
 
 
 
 
 
 
 
 
 
 
109
  # tokenize
110
  for i in $(seq 120 140);
111
  do
speaker_embedding_pyannote.py ADDED
@@ -0,0 +1,36 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ """Pyannote speaker embedding model.
2
+ - pip install pyannote.audio
3
+ - feature dimension: 512
4
+ - source: https://huggingface.co/pyannote/embedding
5
+ """
6
+ from typing import Optional, Union, Tuple
7
+ import torch
8
+ import numpy as np
9
+ from pyannote.audio import Model
10
+ from pyannote.audio import Inference
11
+ from pyannote.audio.core.inference import fix_reproducibility, map_with_specifications
12
+
13
+
14
+ class PyannoteSE:
15
+
16
+ def __init__(self):
17
+ model = Model.from_pretrained("pyannote/embedding")
18
+ self.device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
19
+ model.to(self.device)
20
+ model.eval()
21
+ self.inference = Inference(model, window="whole")
22
+
23
+ def get_speaker_embedding(self, wav: np.ndarray, sampling_rate: Optional[int] = None) -> np.ndarray:
24
+ wav = torch.as_tensor(wav.reshape(1, -1), dtype=torch.float32).to(self.device)
25
+ fix_reproducibility(self.inference.device)
26
+ if self.inference.window == "sliding":
27
+ return self.inference.slide(wav, sampling_rate, hook=None)
28
+
29
+ outputs: Union[np.ndarray, Tuple[np.ndarray]] = self.inference.infer(wav[None])
30
+
31
+ def __first_sample(outputs: np.ndarray, **kwargs) -> np.ndarray:
32
+ return outputs[0]
33
+
34
+ return map_with_specifications(
35
+ self.inference.model.specifications, __first_sample, outputs
36
+ )
tokenize_dataset_s2s.py CHANGED
@@ -54,6 +54,7 @@ dataset = dataset.map(
54
  desc="tokenize dataset"
55
  )
56
  DatasetDict({"train": dataset}).push_to_hub(f"{hf_org}/{hf_dataset}.tokenized", config_name=f"subset_{dataset_id}")
 
57
  cache_dir = f"{expanduser('~')}/.cache/huggingface/datasets/{hf_org}___{hf_dataset}/subset_{dataset_id}"
58
  if os.path.exists(cache_dir):
59
  shutil.rmtree(cache_dir)
 
54
  desc="tokenize dataset"
55
  )
56
  DatasetDict({"train": dataset}).push_to_hub(f"{hf_org}/{hf_dataset}.tokenized", config_name=f"subset_{dataset_id}")
57
+ # DatasetDict({"train": dataset}).push_to_hub(f"{hf_org}/{hf_dataset}.tokenized.encodec", config_name=f"subset_{dataset_id}")
58
  cache_dir = f"{expanduser('~')}/.cache/huggingface/datasets/{hf_org}___{hf_dataset}/subset_{dataset_id}"
59
  if os.path.exists(cache_dir):
60
  shutil.rmtree(cache_dir)