import torch from contants import config from bert_vits2.text.japanese_v200 import text2sep_kata def get_bert_feature(text, word2ph, tokenizer, model, device=config.system.device, **kwargs): sep_text, _, _ = text2sep_kata(text) sep_tokens = [tokenizer.tokenize(t) for t in sep_text] sep_ids = [tokenizer.convert_tokens_to_ids(t) for t in sep_tokens] sep_ids = [2] + [item for sublist in sep_ids for item in sublist] + [3] return get_bert_feature_with_token(sep_ids, word2ph, tokenizer, model, device) def get_bert_feature_with_token(tokens, word2ph, tokenizer, model, device=config.system.device): with torch.no_grad(): inputs = torch.tensor(tokens).to(device).unsqueeze(0) token_type_ids = torch.zeros_like(inputs).to(device) attention_mask = torch.ones_like(inputs).to(device) inputs = { "input_ids": inputs, "token_type_ids": token_type_ids, "attention_mask": attention_mask, } # 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].float().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