|
import torch |
|
from transformers import AutoTokenizer, AutoModelForMaskedLM |
|
|
|
device = torch.device("cuda" if torch.cuda.is_available() else "cpu") |
|
|
|
tokenizer = AutoTokenizer.from_pretrained("./bert/chinese-roberta-wwm-ext-large") |
|
model = AutoModelForMaskedLM.from_pretrained("./bert/chinese-roberta-wwm-ext-large").to(device) |
|
|
|
def get_bert_feature(text, word2ph): |
|
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 len(word2ph) == len(text)+2 |
|
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 |
|
|
|
if __name__ == '__main__': |
|
|
|
import torch |
|
|
|
word_level_feature = torch.rand(38, 1024) |
|
word2phone = [1, 2, 1, 2, 2, 1, 2, 2, 1, 2, 2, 1, 2, 2, 2, 2, 2, 1, 1, 2, 2, 1, 2, 2, 2, 2, 1, 2, 2, 2, 2, 2, 1, 2, 2, 2, 2, 1] |
|
|
|
|
|
total_frames = sum(word2phone) |
|
print(word_level_feature.shape) |
|
print(word2phone) |
|
phone_level_feature = [] |
|
for i in range(len(word2phone)): |
|
print(word_level_feature[i].shape) |
|
|
|
|
|
repeat_feature = word_level_feature[i].repeat(word2phone[i], 1) |
|
phone_level_feature.append(repeat_feature) |
|
|
|
phone_level_feature = torch.cat(phone_level_feature, dim=0) |
|
print(phone_level_feature.shape) |
|
|
|
|