Podcastify / melo /text /chinese_bert.py
mrfakename's picture
Init
4300fed
raw
history blame
2.52 kB
import torch
import sys
from transformers import AutoTokenizer, AutoModelForMaskedLM
# model_id = 'hfl/chinese-roberta-wwm-ext-large'
local_path = "./bert/chinese-roberta-wwm-ext-large"
tokenizers = {}
models = {}
def get_bert_feature(text, word2ph, device=None, model_id='hfl/chinese-roberta-wwm-ext-large'):
if model_id not in models:
models[model_id] = AutoModelForMaskedLM.from_pretrained(
model_id
).to(device)
tokenizers[model_id] = AutoTokenizer.from_pretrained(model_id)
model = models[model_id]
tokenizer = tokenizers[model_id]
if (
sys.platform == "darwin"
and torch.backends.mps.is_available()
and device == "cpu"
):
device = "mps"
if not device:
device = "cuda"
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()
# import pdb; pdb.set_trace()
# 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) # 12个词,每个词1024维特征
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,
]
# 计算总帧数
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)
# 对每个词重复word2phone[i]次
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) # torch.Size([36, 1024])