# -*- coding: UTF-8 -*- #鬼畜音源的活字印刷 #作者：DSP_8192 import soundfile as sf import psola import numpy as np from pypinyin import lazy_pinyin import json from pathlib import Path from io import BytesIO import base64 #-------------------------------------------- #全局变量 #-------------------------------------------- #目标采样率 _targetSR = 44100 #-------------------------------------------- #自定义函数 #-------------------------------------------- #文件路径转文件夹路径 def _fileName2FolderName(fileName): for i in range(len(fileName)-1, -1, -1): if fileName[i] == '\\' or fileName[i] == '/': return fileName[0:i+1] #标准化音频，统一音量 def _normalizeAudio(data): rms = np.sqrt(np.mean(data**2)) normData = data / rms * 0.2 return normData #读取音频文件 def _loadAudio(fileDir, norm): data, sampleRate = sf.read(fileDir) #双声道转单声道 if (len(data.shape) == 2): #左右声道相加除以2 data = (data[:, 0] + data[:, 1]) / 2 #统一采样率 if (sampleRate != _targetSR): #计算转换后的长度 newLength = int((_targetSR / sampleRate) * len(data)) #转换 data = np.interp(np.array(range(newLength)), np.linspace(0,newLength-1,len(data)), data) #标准化 if norm: data = _normalizeAudio(data) return data #改变音高和速度 def _modifyPitchAndSpeed(data, pitchMultiple, speedMultiple): if (pitchMultiple == 1 and speedMultiple == 1): #没有改动的必要，直接返回 return data elif (pitchMultiple > 2 or speedMultiple < 0.5): print("过于极端的音调和速度参数可能导致输出结果与预期不符，故不作改动") return data else: #第一次拉伸 if (speedMultiple / pitchMultiple == 1): #没有拉伸的必要 step1 = data else: #不改变音高的同时在时间上拉伸（PSOLA） #constant_stretch过小会导致bug，因此分两次拉伸 step1 = psola.vocode(data, _targetSR, constant_stretch=1/pitchMultiple) step1 = psola.vocode(step1, _targetSR, constant_stretch=speedMultiple) #第二次拉伸，以改变音高的方式拉伸回来 newLength = int(len(data) / speedMultiple) step2 = np.interp(np.array(range(newLength)), np.linspace(0,newLength-1,len(step1)), step1) return step2 #-------------------------------------------- #活字印刷类 #-------------------------------------------- class huoZiYinShua: def __init__(self, configuration): try: #读取设置文件 dictFile = open(configuration["dictFile"], encoding="utf8") #读取单字词典 (json) ysddTableFile = open(configuration["ysddTableFile"], encoding="utf8") #读取原声大碟文本与文件名对照表 (json) self.__voicePath = configuration["sourceDirectory"] #单字音频文件存放目录 self.__ysddPath = configuration["ysddSourceDirectory"] #原声大碟音频文件存放目录 self.__dictionary = json.load(dictFile) #定义非中文字符读法的词典 self.__ysddTable = json.load(ysddTableFile) #原声大碟文本与文件名对照表 #统一为小写 dictItems = list(self.__ysddTable.items()) #转换为list是为了切断dictItems与self.__ysddTable的联系，否则报错dictionary changed size during iteration for i in dictItems: self.__ysddTable[i[0].lower()] = self.__ysddTable.pop(i[0]) #从长到短排序，越长片段优先级越高 self.__ysddTable = sorted(self.__ysddTable.items(), key=lambda x:len(x[0]), reverse=True) self.__ysddTable = dict(self.__ysddTable) self.__configSucceed = True except Exception as e: self.__configSucceed = False print(e) #配置是否成功 def configSucceed(self): return self.__configSucceed #返回b64音频 def export(self, rawData, inYsddMode=False, pitchMult=1, speedMult=1, norm=False, reverse=False): self.__concatenate(rawData, inYsddMode, pitchMult, speedMult, norm, reverse) base64_data = base64.b64encode(self.__export().getvalue()) return base64_data #生成中间文件 def __concatenate(self, rawData, inYsddMode, pitchMult, speedMult, norm, reverse): missingPinyin = [] self.__concatenated = np.array([]) #预处理，转为小写 rawData = rawData.lower() pronunciations = [] #分割使用活字印刷的部分和使用原声大碟的部分 splitted = [[rawData, False]] #[文本, 是否使用原声大碟] #遍历要匹配的句子 if inYsddMode: for ysdd in self.__ysddTable.items(): #遍历文本 i = -1 while i < (len(splitted) - 1): i += 1 if splitted[i][1]: #已经被划分为原声大碟部分 continue #存在匹配 if ysdd[0] in splitted[i][0]: indexBegin = splitted[i][0].index(ysdd[0]) #获取开始位置 #分割 splitted.insert(i+1, [splitted[i][0][indexBegin:indexBegin+len(ysdd[0])], True]) splitted.insert(i+2, [splitted[i][0][indexBegin+len(ysdd[0]):], False]) splitted[i][0] = splitted[i][0][0:indexBegin] #转换自定义的字符 for i in range(0, len(splitted)): pronunciations.append([]) #使用活字印刷 if (not splitted[i][1]): pronunciations[i].append("") for ch in splitted[i][0]: if ch in self.__dictionary: #词典中存在匹配，转换 pronunciations[i][0] += self.__dictionary[ch] + " " else: #保持不变 pronunciations[i][0] += ch + " " pronunciations[i].append(False) #标记 #使用原声大碟 else: pronunciations[i].append(splitted[i][0]) #直接复制 pronunciations[i].append(True) #标记 #拼接音频 for i in range(0, len(pronunciations)): #使用活字印刷 if (not pronunciations[i][1]): #将汉字转换成拼音 pinyin = lazy_pinyin(pronunciations[i][0]) #拆成单独的字 for text in pinyin: for word in text.split(): #拼接每一个字 try: self.__concatenated = np.concatenate((self.__concatenated, _loadAudio(self.__voicePath + word + ".wav", norm))) #如果出现错误 except Exception as e: if word != "，": print(e) #显示错误信息 #加入缺失素材列表 if word not in missingPinyin: missingPinyin.append(word) #以空白音频代替 self.__concatenated = np.concatenate((self.__concatenated, np.zeros(int(_targetSR/4)))) #使用原声大碟 else: #拼接 try: self.__concatenated = np.concatenate((self.__concatenated, _loadAudio(self.__ysddPath + self.__ysddTable[pronunciations[i][0]], norm))) #如果出现错误 except Exception as e: print(e) #显示错误信息 #加入缺失素材列表 if self.__ysddTable[pronunciations[i][0]] not in missingPinyin: missingPinyin.append(self.__ysddTable[pronunciations[i][0]]) #以空白音频代替 self.__concatenated = np.concatenate((self.__concatenated, np.zeros(int(_targetSR/4)))) #音高偏移 self.__concatenated = _modifyPitchAndSpeed(self.__concatenated, pitchMult, speedMult) #倒放 if(reverse): self.__concatenated = np.flip(self.__concatenated) #如果缺失拼音，则发出警告 #if len(missingPinyin) != 0: # print("警告：缺失或未定义{}".format(missingPinyin)) #返回二进制音频 def __export(self): bytesout = BytesIO() sf.write(bytesout, self.__concatenated, _targetSR, format="MP3") return bytesout