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])