tuna2134/sbv2-v2
1
0
Fork 0
Code Issues Pull Requests 1 Actions Packages Projects Releases Wiki Activity

Update japanese_bert.py

Browse Source
This commit is contained in:
Stardust·减
2023-09-23 20:12:08 +08:00
committed by GitHub
parent 4e549c5d49
commit 657c835ffd
1 changed files with 9 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

13
text/japanese_bert.py
View File

@@ -2,7 +2,12 @@ import torch
from transformers import AutoTokenizer, AutoModelForMaskedLM
import sys
tokenizer = AutoTokenizer.from_pretrained("./bert/bert-base-japanese-v3")
BERT = './bert/bert-large-japanese-v2'
tokenizer = AutoTokenizer.from_pretrained(BERT)
# bert-large model has 25 hidden layers.You can decide which layer to use by setting this variable to a specific value
# default value is 3(untested)
BERT_LAYER = 3
def get_bert_feature(text, word2ph, device=None):
@@ -14,7 +19,7 @@ def get_bert_feature(text, word2ph, device=None):
device = "mps"
if not device:
device = "cuda"
model = AutoModelForMaskedLM.from_pretrained("./bert/bert-base-japanese-v3").to(
model = AutoModelForMaskedLM.from_pretrained(BERT).to(
device
)
with torch.no_grad():
@@ -22,12 +27,12 @@ def get_bert_feature(text, word2ph, device=None):
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()
res = res['hidden_states'][BERT_LAYER]
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)
repeat_feature = res[0][i].repeat(word2phone[i], 1)
phone_level_feature.append(repeat_feature)
phone_level_feature = torch.cat(phone_level_feature, dim=0)