28 lines
1023 B
Python
28 lines
1023 B
Python
import torch
|
|
from transformers import AutoTokenizer, AutoModelForMaskedLM
|
|
|
|
|
|
tokenizer = AutoTokenizer.from_pretrained("./bert/bert-base-japanese-v3")
|
|
|
|
def get_bert_feature(text, word2ph):
|
|
if sys.platform == "darwin" and torch.backends.mps.is_available() and device == "cpu"
|
|
device == "mps"
|
|
model = AutoModelForMaskedLM.from_pretrained("./bert/bert-base-japanese-v3").to(device)
|
|
with torch.no_grad():
|
|
inputs = tokenizer(text, return_tensors='pt')
|
|
for i in inputs:
|
|
inputs[i] = inputs[i].to(device)
|
|
res = model(**inputs, output_hidden_states=True)
|
|
res = torch.cat(res['hidden_states'][-3:-2], -1)[0].cpu()
|
|
assert inputs['input_ids'].shape[-1] == len(word2ph)
|
|
word2phone = word2ph
|
|
phone_level_feature = []
|
|
for i in range(len(word2phone)):
|
|
repeat_feature = res[i].repeat(word2phone[i], 1)
|
|
phone_level_feature.append(repeat_feature)
|
|
|
|
phone_level_feature = torch.cat(phone_level_feature, dim=0)
|
|
|
|
|
|
return phone_level_feature.T
|