PaddleSpeech/docs/tutorial/tts/tts_tutorial.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# 『听』和『说』\n",
    "人类通过听觉获取的信息大约占所有感知信息的 20% ~ 30%。声音存储了丰富的语义以及时序信息，由专门负责听觉的器官接收信号，产生一系列连锁刺激后，在人类大脑的皮层听区进行处理分析，获取语义和知识。近年来，随着深度学习算法上的进步以及不断丰厚的硬件资源条件，**文本转语音（Text-to-Speech, TTS）** 技术在移动、虚拟娱乐等领域得到了广泛的应用。</font>\n",
    "## \"听\"书\n",
    "使用 [PaddleOCR](https://github.com/PaddlePaddle/PaddleOCR) 直接获取书籍上的文字。"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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
      "text/plain": [
       "<PIL.JpegImagePlugin.JpegImageFile image mode=RGB size=1200x462 at 0x7F7CC46D2710>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "from PIL import Image\n",
    "img_path = 'source/ocr_result.jpg'\n",
    "im = Image.open(img_path)\n",
    "im.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "使用 [PaddleSpeech](https://github.com/PaddlePaddle/PaddleSpeech)，阅读上一步识别出来的文字。"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "\n",
       "                <audio controls=\"controls\" >\n",
       "                    <source src=\"data:audio/x-wav;base64,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
       "                    Your browser does not support the audio element.\n",
       "                </audio>\n",
       "              "
      ],
      "text/plain": [
       "<IPython.lib.display.Audio object>"
      ]
     },
     "execution_count": 2,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "import IPython.display as dp\n",
    "dp.Audio(\"source/ocr.wav\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "具体实现代码详见 [Story Talker](https://github.com/DeepSpeech/demos/story_talker/run.sh)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 偶像开口说话\n",
    "*元宇宙来袭，构造你的虚拟人！* 看看 [PaddleGAN](https://github.com/PaddlePaddle/PaddleGAN) 怎样合成唇形，让WiFi之母——海蒂·拉玛说话。"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "\n",
       "<video controls width=\"600\" height=\"360\" src=\"source/tts_lips.mp4\">animation</video>\n"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "from IPython.display import HTML\n",
    "html_str = '''\n",
    "<video controls width=\"600\" height=\"360\" src=\"{}\">animation</video>\n",
    "'''.format(\"source/tts_lips.mp4\")\n",
    "dp.display(HTML(html_str))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "具体实现代码请参考 [Metaverse](https://github.com/DeepSpeech/demos/metaverse/run.sh)。\n",
    "\n",
    "下面让我们来系统地学习语音方面的知识，看看怎样使用 **PaddleSpeech** 实现基本的语音功能，以及怎样结合光学字符识别（Optical Character Recognition，OCR）、自然语言处理（Natural Language Processing，NLP）等技术“听”书、让名人开口说话。"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# 前言\n",
    "\n",
    "## 背景知识\n",
    "为了更好地了解文本转语音任务的要素，我们先简要地回顾一下文本转语音的发展历史。如果你对此已经有所了解，或希望能尽快使用代码实现，请直接跳至第四章[实践](#实践)。\n",
    "### 定义\n",
    "<!----\n",
    "Note: \n",
    "1.此句抄自 [李沐Dive into Dive Learning](https://zh-v2.d2l.ai/chapter_introduction/index.html)\n",
    "2.修改参考A survey on Neural Speech Sysnthesis.\n",
    "---> \n",
    "文本转语音，又称语音合成（Speech Sysnthesis），指的是将一段文本按照一定需求转化成对应的音频，这种特性决定了的输出数据比输入输入长得多。文本转语音是一项包含了语义学、声学、数字信号处理以及机器学习的等多项学科的交叉任务。虽然辨识低质量音频文件的内容对人类来说很容易，但这对计算机来说并非易事。\n",
    "\n",
    "按照不同的应用需求，更广义的语音合成研究包括：*语音转换*，例如说话人转换、语音到歌唱转换、语音情感转换、口音转换等；*歌唱合成*，例如歌词到歌唱转换、可视语音合成等。\n",
    "\n",
    "### 发展历史\n",
    "\n",
    "<!--\n",
    "以下摘自维基百科 https://en.wikipedia.org/wiki/Speech_synthesis\n",
    "--->\n",
    "\n",
    "在第二次工业革命之前，语音的合成主要以机械式的音素合成为主。1779年，德裔丹麦科学家 Christian Gottlieb Kratzenstein 建造了人类的声道模型，使其可以产生五个长元音。1791年， Wolfgang von Kempelen 添加了唇和舌的模型，使其能够发出辅音和元音。贝尔实验室于20世纪30年代发明了声码器（Vocoder），将语音自动分解为音调和共振，此项技术由 Homer Dudley 改进为键盘式合成器并于 1939年纽约世界博览会展出。\n",
    "\n",
    "第一台基于计算机的语音合成系统起源于20世纪50年代。1961年，IBM 的 John Larry Kelly，以及 Louis Gerstman 使用 IBM 704 计算机合成语音，成为贝尔实验室最著名的成就之一。1975年，第一代语音合成系统之一 —— MUSA（MUltichannel Speaking Automation）问世，其由一个独立的硬件和配套的软件组成。1978年发行的第二个版本也可以进行无伴奏演唱。90 年代的主流是采用 MIT 和贝尔实验室的系统，并结合自然语言处理模型。\n",
    "![语音合成技术的发展历史](./source/tts-timeline.png)\n",
    "\n",
    "### 主流方法\n",
    "\n",
    "当前的主流方法分为**基于统计参数的语音合成**、**波形拼接语音合成**、**混合方法**以及**端到端神经网络语音合成**。基于参数的语音合成包含隐马尔可夫模型（Hidden Markov Model,HMM）以及深度学习网络（Deep Neural Network，DNN）。端到端的方法保函声学模型+声码器以及“完全”端到端方法。\n",
    "\n",
    "\n",
    "## 基于深度学习的语音合成技术\n",
    "\n",
    "### 语音合成基本知识\n",
    "\n",
    "![信号处理流水线](source/signal_pipeline.png)\n",
    "\n",
    "语音合成流水线包含 <font color=\"#ff0000\">**文本前端（Text Frontend）**</font> 、<font color=\"#ff0000\">**声学模型（Acoustic Model）**</font> 和 <font color=\"#ff0000\">**声码器（Vocoder）**</font> 三个主要模块:\n",
    "- 通过文本前端模块将原始文本转换为字符/音素。\n",
    "- 通过声学模型将字符/音素转换为声学特征，如线性频谱图、mel 频谱图、LPC 特征等。\n",
    "- 通过声码器将声学特征转换为波形。\n",
    "\n",
    "<img style=\"float: center;\" src=\"source/tts_pipeline.png\" width=\"85%\"/>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# 实践\n",
    "<br></br>\n",
    "环境安装请参考 [install](https://github.com/PaddlePaddle/PaddleSpeech/blob/develop/docs/source/install.md)安装教程。 \n",
    "下面使用 **PaddleSpeech** 提供的预训练模型合成中文语音。"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 数据及模型准备"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 获取 PaddlePaddle 预训练模型"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "scrolled": true
   },
   "outputs": [],
   "source": [
    "!mkdir download\n",
    "!wget -P download https://paddlespeech.bj.bcebos.com/Parakeet/released_models/pwgan/pwg_baker_ckpt_0.4.zip\n",
    "!unzip -d download download/pwg_baker_ckpt_0.4.zip\n",
    "!wget -P download https://paddlespeech.bj.bcebos.com/Parakeet/released_models/fastspeech2/fastspeech2_nosil_baker_ckpt_0.4.zip\n",
    "!unzip -d download download/fastspeech2_nosil_baker_ckpt_0.4.zip"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\u001b[01;34mdownload/pwg_baker_ckpt_0.4\u001b[00m\n",
      "|-- pwg_default.yaml\n",
      "|-- pwg_snapshot_iter_400000.pdz\n",
      "`-- pwg_stats.npy\n",
      "\n",
      "0 directories, 3 files\n",
      "\u001b[01;34mdownload/fastspeech2_nosil_baker_ckpt_0.4\u001b[00m\n",
      "|-- default.yaml\n",
      "|-- energy_stats.npy\n",
      "|-- phone_id_map.txt\n",
      "|-- pitch_stats.npy\n",
      "|-- snapshot_iter_76000.pdz\n",
      "`-- speech_stats.npy\n",
      "\n",
      "0 directories, 6 files\n"
     ]
    }
   ],
   "source": [
    "!tree download/pwg_baker_ckpt_0.4\n",
    "!tree download/fastspeech2_nosil_baker_ckpt_0.4"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 导入 Python 包"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "The autoreload extension is already loaded. To reload it, use:\n",
      "  %reload_ext autoreload\n"
     ]
    }
   ],
   "source": [
    "%load_ext autoreload\n",
    "%autoreload 2\n",
    "\n",
    "# 设置 gpu 环境\n",
    "%env CUDA_VISIBLE_DEVICES=0\n",
    "\n",
    "import logging\n",
    "import sys\n",
    "import warnings\n",
    "warnings.filterwarnings('ignore')\n",
    "\n",
    "# 需要将PaddleSpeech项目的根目录放到Python路径中\n",
    "sys.path.insert(0,\"../../../\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {
    "scrolled": true
   },
   "outputs": [],
   "source": [
    "import argparse\n",
    "import os\n",
    "from pathlib import Path\n",
    "import IPython.display as dp\n",
    "import matplotlib.pyplot as plt\n",
    "import numpy as np\n",
    "import paddle\n",
    "import soundfile as sf\n",
    "import yaml\n",
    "from paddlespeech.t2s.frontend.zh_frontend import Frontend\n",
    "from paddlespeech.t2s.models.fastspeech2 import FastSpeech2\n",
    "from paddlespeech.t2s.models.fastspeech2 import FastSpeech2Inference\n",
    "from paddlespeech.t2s.models.parallel_wavegan import PWGGenerator\n",
    "from paddlespeech.t2s.models.parallel_wavegan import PWGInference\n",
    "from paddlespeech.t2s.modules.normalizer import ZScore\n",
    "from yacs.config import CfgNode"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 设置预训练模型的路径"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "========Config========\n",
      "batch_size: 64\n",
      "f0max: 400\n",
      "f0min: 80\n",
      "fmax: 7600\n",
      "fmin: 80\n",
      "fs: 24000\n",
      "max_epoch: 1000\n",
      "model:\n",
      "  adim: 384\n",
      "  aheads: 2\n",
      "  decoder_normalize_before: True\n",
      "  dlayers: 4\n",
      "  dunits: 1536\n",
      "  duration_predictor_chans: 256\n",
      "  duration_predictor_kernel_size: 3\n",
      "  duration_predictor_layers: 2\n",
      "  elayers: 4\n",
      "  encoder_normalize_before: True\n",
      "  energy_embed_dropout: 0.0\n",
      "  energy_embed_kernel_size: 1\n",
      "  energy_predictor_chans: 256\n",
      "  energy_predictor_dropout: 0.5\n",
      "  energy_predictor_kernel_size: 3\n",
      "  energy_predictor_layers: 2\n",
      "  eunits: 1536\n",
      "  init_dec_alpha: 1.0\n",
      "  init_enc_alpha: 1.0\n",
      "  init_type: xavier_uniform\n",
      "  pitch_embed_dropout: 0.0\n",
      "  pitch_embed_kernel_size: 1\n",
      "  pitch_predictor_chans: 256\n",
      "  pitch_predictor_dropout: 0.5\n",
      "  pitch_predictor_kernel_size: 5\n",
      "  pitch_predictor_layers: 5\n",
      "  positionwise_conv_kernel_size: 3\n",
      "  positionwise_layer_type: conv1d\n",
      "  postnet_chans: 256\n",
      "  postnet_filts: 5\n",
      "  postnet_layers: 5\n",
      "  reduction_factor: 1\n",
      "  stop_gradient_from_energy_predictor: False\n",
      "  stop_gradient_from_pitch_predictor: True\n",
      "  transformer_dec_attn_dropout_rate: 0.2\n",
      "  transformer_dec_dropout_rate: 0.2\n",
      "  transformer_dec_positional_dropout_rate: 0.2\n",
      "  transformer_enc_attn_dropout_rate: 0.2\n",
      "  transformer_enc_dropout_rate: 0.2\n",
      "  transformer_enc_positional_dropout_rate: 0.2\n",
      "  use_masking: True\n",
      "  use_scaled_pos_enc: True\n",
      "n_fft: 2048\n",
      "n_mels: 80\n",
      "n_shift: 300\n",
      "num_snapshots: 5\n",
      "num_workers: 4\n",
      "optimizer:\n",
      "  learning_rate: 0.001\n",
      "  optim: adam\n",
      "seed: 10086\n",
      "updater:\n",
      "  use_masking: True\n",
      "win_length: 1200\n",
      "window: hann\n",
      "---------------------\n",
      "allow_cache: True\n",
      "batch_max_steps: 25500\n",
      "batch_size: 6\n",
      "discriminator_grad_norm: 1\n",
      "discriminator_optimizer_params:\n",
      "  epsilon: 1e-06\n",
      "  weight_decay: 0.0\n",
      "discriminator_params:\n",
      "  bias: True\n",
      "  conv_channels: 64\n",
      "  in_channels: 1\n",
      "  kernel_size: 3\n",
      "  layers: 10\n",
      "  nonlinear_activation: LeakyReLU\n",
      "  nonlinear_activation_params:\n",
      "    negative_slope: 0.2\n",
      "  out_channels: 1\n",
      "  use_weight_norm: True\n",
      "discriminator_scheduler_params:\n",
      "  gamma: 0.5\n",
      "  learning_rate: 5e-05\n",
      "  step_size: 200000\n",
      "discriminator_train_start_steps: 100000\n",
      "eval_interval_steps: 1000\n",
      "fmax: 7600\n",
      "fmin: 80\n",
      "fs: 24000\n",
      "generator_grad_norm: 10\n",
      "generator_optimizer_params:\n",
      "  epsilon: 1e-06\n",
      "  weight_decay: 0.0\n",
      "generator_params:\n",
      "  aux_channels: 80\n",
      "  aux_context_window: 2\n",
      "  bias: True\n",
      "  dropout: 0.0\n",
      "  freq_axis_kernel_size: 1\n",
      "  gate_channels: 128\n",
      "  in_channels: 1\n",
      "  interpolate_mode: nearest\n",
      "  kernel_size: 3\n",
      "  layers: 30\n",
      "  nonlinear_activation: None\n",
      "  nonlinear_activation_params:\n",
      "    \n",
      "  out_channels: 1\n",
      "  residual_channels: 64\n",
      "  skip_channels: 64\n",
      "  stacks: 3\n",
      "  upsample_scales: [4, 5, 3, 5]\n",
      "  use_causal_conv: False\n",
      "  use_weight_norm: True\n",
      "generator_scheduler_params:\n",
      "  gamma: 0.5\n",
      "  learning_rate: 0.0001\n",
      "  step_size: 200000\n",
      "lambda_adv: 4.0\n",
      "n_fft: 2048\n",
      "n_mels: 80\n",
      "n_shift: 300\n",
      "num_save_intermediate_results: 4\n",
      "num_snapshots: 10\n",
      "num_workers: 4\n",
      "pin_memory: True\n",
      "remove_short_samples: True\n",
      "save_interval_steps: 5000\n",
      "seed: 42\n",
      "stft_loss_params:\n",
      "  fft_sizes: [1024, 2048, 512]\n",
      "  hop_sizes: [120, 240, 50]\n",
      "  win_lengths: [600, 1200, 240]\n",
      "  window: hann\n",
      "top_db: 60\n",
      "train_max_steps: 400000\n",
      "trim_frame_length: 2048\n",
      "trim_hop_length: 512\n",
      "trim_silence: False\n",
      "win_length: 1200\n",
      "window: hann\n"
     ]
    }
   ],
   "source": [
    "fastspeech2_config = \"download/fastspeech2_nosil_baker_ckpt_0.4/default.yaml\"\n",
    "fastspeech2_checkpoint = \"download/fastspeech2_nosil_baker_ckpt_0.4/snapshot_iter_76000.pdz\"\n",
    "fastspeech2_stat = \"download/fastspeech2_nosil_baker_ckpt_0.4/speech_stats.npy\"\n",
    "pwg_config = \"download/pwg_baker_ckpt_0.4/pwg_default.yaml\"\n",
    "pwg_checkpoint = \"download/pwg_baker_ckpt_0.4/pwg_snapshot_iter_400000.pdz\"\n",
    "pwg_stat = \"download/pwg_baker_ckpt_0.4/pwg_stats.npy\"\n",
    "phones_dict = \"download/fastspeech2_nosil_baker_ckpt_0.4/phone_id_map.txt\"\n",
    "# 读取 conf 配置文件并结构化\n",
    "with open(fastspeech2_config) as f:\n",
    "    fastspeech2_config = CfgNode(yaml.safe_load(f))\n",
    "with open(pwg_config) as f:\n",
    "    pwg_config = CfgNode(yaml.safe_load(f))\n",
    "print(\"========Config========\")\n",
    "print(fastspeech2_config)\n",
    "print(\"---------------------\")\n",
    "print(pwg_config)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 文本前端（Text Frontend）\n",
    "\n",
    "\n",
    "一个文本前端模块主要包含:\n",
    "- 分段（Text Segmentation）\n",
    "- 文本正则化（Text Normalization, TN）\n",
    "- 分词（Word Segmentation, 主要是在中文中）\n",
    "- 词性标注（Part-of-Speech, PoS）\n",
    "- 韵律预测（Prosody）\n",
    "- 字音转换（Grapheme-to-Phoneme，G2P）\n",
    "<br></br>\n",
    "<font size=2>（Grapheme: **语言**书写系统的最小有意义单位; Phoneme: 区分单词的最小**语音**单位）</font>\n",
    "    - 多音字（Polyphone）\n",
    "    - 变调（Tone Sandhi）\n",
    "        - “一”、“不”变调\n",
    "        - 三声变调\n",
    "        - 轻声变调\n",
    "        - 儿化音\n",
    "        - 方言\n",
    "- ...\n",
    "<br></br>\n",
    "\n",
    "（输入给声学模型之前，还需要把音素序列转换为 id）\n",
    "\n",
    "\n",
    "其中最重要的模块是<font color=\"#ff0000\"> 文本正则化 </font>模块和<font color=\"#ff0000\"> 字音转换（TTS 中更常用 G2P 代指） </font>模块。\n",
    "\n",
    "\n",
    "各模块输出示例:\n",
    "```text\n",
    "• Text: 全国一共有112所211高校\n",
    "• Text Normalization: 全国一共有一百一十二所二一一高校\n",
    "• Word Segmentation: 全国/一共/有/一百一十二/所/二一一/高校/\n",
    "• G2P（注意此句中“一”的读音）:\n",
    "    quan2 guo2 yi2 gong4 you3 yi4 bai3 yi1 shi2 er4 suo3 er4 yao1 yao1 gao1 xiao4\n",
    "    （可以进一步把声母和韵母分开）\n",
    "    q uan2 g uo2 y i2 g ong4 y ou3 y i4 b ai3 y i1 sh i2 er4 s uo3 er4 y ao1 y ao1 g ao1 x iao4\n",
    "    （把音调和声韵母分开）\n",
    "    q uan g uo y i g ong y ou y i b ai y i sh i er s uo er y ao y ao g ao x iao\n",
    "    0 2 0 2 0 2 0 4 0 3 ...\n",
    "• Prosody (prosodic words #1, prosodic phrases #2, intonation phrases #3, sentence #4):\n",
    "    全国#2一共有#2一百#1一十二所#2二一一#1高校#4\n",
    "    （分词的结果一般是固定的，但是不同人习惯不同，可能有不同的韵律）\n",
    "```\n",
    "\n",
    "<br></br>\n",
    "文本前端模块的设计需要结合很多专业的语义学知识和经验。人类在读文本的时候可以自然而然地读出正确的发音，但是这些先验知识计算机并不知晓。\n",
    "例如，对于一个句子的分词：\n",
    "\n",
    "```text\n",
    "我也想过过过儿过过的生活\n",
    "我也想/过过/过儿/过过的/生活\n",
    "\n",
    "货拉拉拉不拉拉布拉多\n",
    "货拉拉/拉不拉/拉布拉多\n",
    "\n",
    "南京市长江大桥\n",
    "南京市长/江大桥\n",
    "南京市/长江大桥\n",
    "```\n",
    "或者是词的变调和儿化音：\n",
    "```\n",
    "你要不要和我们一起出去玩？\n",
    "你要不（2声）要和我们一（4声）起出去玩（儿）？\n",
    "\n",
    "不好，我要一个人出去。\n",
    "不（4声）好，我要一（2声）个人出去。\n",
    "\n",
    "（以下每个词的所有字都是三声的，请你读一读，体会一下在读的时候，是否每个字都被读成了三声？）\n",
    "纸老虎、虎骨酒、展览馆、岂有此理、手表厂有五种好产品\n",
    "```\n",
    "又或是多音字，这类情况通常需要先正确分词：\n",
    "```text\n",
    "人要行，干一行行一行，一行行行行行;\n",
    "人要是不行，干一行不行一行，一行不行行行不行。\n",
    "\n",
    "佟大为妻子产下一女\n",
    "\n",
    "海水朝朝朝朝朝朝朝落\n",
    "浮云长长长长长长长消\n",
    "```\n",
    "\n",
    "PaddleSpeech Text-to-Speech的文本前端解决方案:\n",
    "- [文本正则](https://github.com/PaddlePaddle/PaddleSpeech/tree/develop/examples/other/tn)\n",
    "- [G2P](https://github.com/PaddlePaddle/PaddleSpeech/tree/develop/examples/other/g2p):\n",
    "    - 多音字模块: pypinyin/g2pM\n",
    "    - 变调模块: 用分词 + 规则\n",
    "<br></br>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 构造文本前端对象\n",
    "<font size=4>传入`phones_dict`，把相应的`phones`转换成`phone_ids`。</font>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {
    "scrolled": true
   },
   "outputs": [],
   "source": [
    "# 传入 phones_dict 会把相应的 phones 转换成 phone_ids\n",
    "frontend = Frontend(phone_vocab_path=phones_dict)\n",
    "print(\"Frontend done!\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 调用文本前端\n",
    "\n",
    "文本前端对输入的数据进行正则化的时候会进行分句，你也可以将以下`input`替换成`\"我每天中午12:00起床\"`或者`\"我出生于2005/11/08，那天的最低气温达到-10°C\"`。\n",
    "若`merge_sentences`设置为`False`，则多个子句并行调用声学模型和声码器提升合成速度；若`merge_sentences`设置为`True`，`input_ids[\"phone_ids\"][0]`则表示整句的`phone_ids`。"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Building prefix dict from the default dictionary ...\n",
      "DEBUG:jieba:Building prefix dict from the default dictionary ...\n",
      "Loading model from cache /tmp/jieba.cache\n",
      "DEBUG:jieba:Loading model from cache /tmp/jieba.cache\n",
      "Loading model cost 5.331 seconds.\n",
      "DEBUG:jieba:Loading model cost 5.331 seconds.\n",
      "Prefix dict has been built successfully.\n",
      "DEBUG:jieba:Prefix dict has been built successfully.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "----------------------------\n",
      "text norm results:\n",
      "['你好，', '欢迎使用百度飞桨框架进行深度学习研究！']\n",
      "----------------------------\n",
      "g2p results:\n",
      "[['n', 'i2', 'h', 'ao3', 'sp', 'h', 'uan1', 'ing2', 'sh', 'iii3', 'iong4', 'b', 'ai3', 'd', 'u4', 'f', 'ei1', 'j', 'iang3', 'k', 'uang1', 'j', 'ia4', 'j', 'in4', 'x', 'ing2', 'sh', 'en1', 'd', 'u4', 'x', 've2', 'x', 'i2', 'ian2', 'j', 'iou1', 'sp']]\n",
      "----------------------------\n",
      "phone_ids:Tensor(shape=[39], dtype=int64, place=CUDAPlace(0), stop_gradient=True,\n",
      "       [155, 73 , 71 , 29 , 179, 71 , 199, 126, 177, 115, 138, 37 , 9  , 40 ,\n",
      "        186, 69 , 46 , 151, 89 , 152, 204, 151, 80 , 151, 123, 260, 126, 177,\n",
      "        51 , 40 , 186, 260, 251, 260, 73 , 83 , 151, 140, 179])\n"
     ]
    }
   ],
   "source": [
    "input = \"你好，欢迎使用百度飞桨框架进行深度学习研究！\"\n",
    "input_ids = frontend.get_input_ids(input, merge_sentences=True, print_info=True)\n",
    "phone_ids = input_ids[\"phone_ids\"][0]\n",
    "print(\"phone_ids:%s\"%phone_ids)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 用深度学习实现文本前端\n",
    "<img style=\"float: center;\" src=\"source/text_frontend_struct.png\" width=\"100%\"/>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 声学模型（Acoustic Model）\n",
    "\n",
    "声学模型将字符/音素转换为声学特征，如线性频谱图、mel 频谱图、LPC 特征等，声学特征以 “帧” 为单位，一般一帧是 10ms 左右，一个音素一般对应 5~20 帧左右, 声学模型需要解决的是 <font color=\"#ff0000\">“不等长序列间的映射问题”</font>，“不等长”是指，同一个人发不同音素的持续时间不同，同一个人在不同时刻说同一句话的语速可能不同，对应各个音素的持续时间不同，不同人说话的特色不同，对应各个音素的持续时间不同。这是一个困难的“一对多”问题。\n",
    "```\n",
    "# 卡尔普陪外孙玩滑梯\n",
    "000001|baker_corpus|sil 20 k 12 a2 4 er2 10 p 12 u3 12 p 9 ei2 9 uai4 15 s 11 uen1 12 uan2 14 h 10 ua2 11 t 15 i1 16 sil 20\n",
    "```\n",
    "\n",
    "声学模型主要分为自回归模型和非自回归模型，其中自回归模型在 `t` 时刻的预测需要依赖 `t-1` 时刻的输出作为输入，预测时间长，但是音质相对较好，非自回归模型不存在预测上的依赖关系，预测时间快，音质相对较差。\n",
    "\n",
    "<br></br>\n",
    "主流声学模型发展的脉络:\n",
    "- 自回归模型:\n",
    "    - Tacotron\n",
    "    - Tacotron2\n",
    "    - Transformer TTS\n",
    "- 非自回归模型:\n",
    "    - FastSpeech\n",
    "    - SpeedySpeech\n",
    "    - FastPitch\n",
    "    - FastSpeech2\n",
    "    - ...\n",
    " \n",
    "<br></br>\n",
    "在本教程中，我们使用 `FastSpeech2` 作为声学模型。\n",
    "![FastSpeech2](source/fastspeech2.png)\n",
    "PaddleSpeech TTS 实现的 FastSpeech2 与论文不同的地方在于，我们使用的的是 phone 级别的 `pitch` 和 `energy`(与 FastPitch 类似)。\n",
    "![FastPitch](source/fastpitch.png)\n",
    "更多关于[声学模型的发展及改进](https://paddlespeech.readthedocs.io/en/latest/tts/models_introduction.html)。"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 初始化声学模型 FastSpeech2"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "vocab_size: 268\n",
      "None\n",
      "FastSpeech2Inference(\n",
      "  (normalizer): ZScore()\n",
      "  (acoustic_model): FastSpeech2(\n",
      "    (encoder): Encoder(\n",
      "      (embed): Sequential(\n",
      "        (0): Embedding(268, 384, padding_idx=0, sparse=False)\n",
      "        (1): ScaledPositionalEncoding(\n",
      "          (dropout): Dropout(p=0.2, axis=None, mode=upscale_in_train)\n",
      "        )\n",
      "      )\n",
      "      (encoders): MultiSequential(\n",
      "        (0): EncoderLayer(\n",
      "          (self_attn): MultiHeadedAttention(\n",
      "            (linear_q): Linear(in_features=384, out_features=384, dtype=float32)\n",
      "            (linear_k): Linear(in_features=384, out_features=384, dtype=float32)\n",
      "            (linear_v): Linear(in_features=384, out_features=384, dtype=float32)\n",
      "            (linear_out): Linear(in_features=384, out_features=384, dtype=float32)\n",
      "            (dropout): Dropout(p=0.2, axis=None, mode=upscale_in_train)\n",
      "          )\n",
      "          (feed_forward): MultiLayeredConv1d(\n",
      "            (w_1): Conv1D(384, 1536, kernel_size=[3], padding=1, data_format=NCL)\n",
      "            (w_2): Conv1D(1536, 384, kernel_size=[3], padding=1, data_format=NCL)\n",
      "            (dropout): Dropout(p=0.2, axis=None, mode=upscale_in_train)\n",
      "            (relu): ReLU()\n",
      "          )\n",
      "          (norm1): LayerNorm(normalized_shape=[384], epsilon=1e-05)\n",
      "          (norm2): LayerNorm(normalized_shape=[384], epsilon=1e-05)\n",
      "          (dropout): Dropout(p=0.2, axis=None, mode=upscale_in_train)\n",
      "        )\n",
      "        (1): EncoderLayer(\n",
      "          (self_attn): MultiHeadedAttention(\n",
      "            (linear_q): Linear(in_features=384, out_features=384, dtype=float32)\n",
      "            (linear_k): Linear(in_features=384, out_features=384, dtype=float32)\n",
      "            (linear_v): Linear(in_features=384, out_features=384, dtype=float32)\n",
      "            (linear_out): Linear(in_features=384, out_features=384, dtype=float32)\n",
      "            (dropout): Dropout(p=0.2, axis=None, mode=upscale_in_train)\n",
      "          )\n",
      "          (feed_forward): MultiLayeredConv1d(\n",
      "            (w_1): Conv1D(384, 1536, kernel_size=[3], padding=1, data_format=NCL)\n",
      "            (w_2): Conv1D(1536, 384, kernel_size=[3], padding=1, data_format=NCL)\n",
      "            (dropout): Dropout(p=0.2, axis=None, mode=upscale_in_train)\n",
      "            (relu): ReLU()\n",
      "          )\n",
      "          (norm1): LayerNorm(normalized_shape=[384], epsilon=1e-05)\n",
      "          (norm2): LayerNorm(normalized_shape=[384], epsilon=1e-05)\n",
      "          (dropout): Dropout(p=0.2, axis=None, mode=upscale_in_train)\n",
      "        )\n",
      "        (2): EncoderLayer(\n",
      "          (self_attn): MultiHeadedAttention(\n",
      "            (linear_q): Linear(in_features=384, out_features=384, dtype=float32)\n",
      "            (linear_k): Linear(in_features=384, out_features=384, dtype=float32)\n",
      "            (linear_v): Linear(in_features=384, out_features=384, dtype=float32)\n",
      "            (linear_out): Linear(in_features=384, out_features=384, dtype=float32)\n",
      "            (dropout): Dropout(p=0.2, axis=None, mode=upscale_in_train)\n",
      "          )\n",
      "          (feed_forward): MultiLayeredConv1d(\n",
      "            (w_1): Conv1D(384, 1536, kernel_size=[3], padding=1, data_format=NCL)\n",
      "            (w_2): Conv1D(1536, 384, kernel_size=[3], padding=1, data_format=NCL)\n",
      "            (dropout): Dropout(p=0.2, axis=None, mode=upscale_in_train)\n",
      "            (relu): ReLU()\n",
      "          )\n",
      "          (norm1): LayerNorm(normalized_shape=[384], epsilon=1e-05)\n",
      "          (norm2): LayerNorm(normalized_shape=[384], epsilon=1e-05)\n",
      "          (dropout): Dropout(p=0.2, axis=None, mode=upscale_in_train)\n",
      "        )\n",
      "        (3): EncoderLayer(\n",
      "          (self_attn): MultiHeadedAttention(\n",
      "            (linear_q): Linear(in_features=384, out_features=384, dtype=float32)\n",
      "            (linear_k): Linear(in_features=384, out_features=384, dtype=float32)\n",
      "            (linear_v): Linear(in_features=384, out_features=384, dtype=float32)\n",
      "            (linear_out): Linear(in_features=384, out_features=384, dtype=float32)\n",
      "            (dropout): Dropout(p=0.2, axis=None, mode=upscale_in_train)\n",
      "          )\n",
      "          (feed_forward): MultiLayeredConv1d(\n",
      "            (w_1): Conv1D(384, 1536, kernel_size=[3], padding=1, data_format=NCL)\n",
      "            (w_2): Conv1D(1536, 384, kernel_size=[3], padding=1, data_format=NCL)\n",
      "            (dropout): Dropout(p=0.2, axis=None, mode=upscale_in_train)\n",
      "            (relu): ReLU()\n",
      "          )\n",
      "          (norm1): LayerNorm(normalized_shape=[384], epsilon=1e-05)\n",
      "          (norm2): LayerNorm(normalized_shape=[384], epsilon=1e-05)\n",
      "          (dropout): Dropout(p=0.2, axis=None, mode=upscale_in_train)\n",
      "        )\n",
      "      )\n",
      "      (after_norm): LayerNorm(normalized_shape=[384], epsilon=1e-05)\n",
      "    )\n",
      "    (duration_predictor): DurationPredictor(\n",
      "      (conv): LayerList(\n",
      "        (0): Sequential(\n",
      "          (0): Conv1D(384, 256, kernel_size=[3], padding=1, data_format=NCL)\n",
      "          (1): ReLU()\n",
      "          (2): LayerNorm(normalized_shape=[256], epsilon=1e-05)\n",
      "          (3): Dropout(p=0.1, axis=None, mode=upscale_in_train)\n",
      "        )\n",
      "        (1): Sequential(\n",
      "          (0): Conv1D(256, 256, kernel_size=[3], padding=1, data_format=NCL)\n",
      "          (1): ReLU()\n",
      "          (2): LayerNorm(normalized_shape=[256], epsilon=1e-05)\n",
      "          (3): Dropout(p=0.1, axis=None, mode=upscale_in_train)\n",
      "        )\n",
      "      )\n",
      "      (linear): Linear(in_features=256, out_features=1, dtype=float32)\n",
      "    )\n",
      "    (pitch_predictor): VariancePredictor(\n",
      "      (conv): LayerList(\n",
      "        (0): Sequential(\n",
      "          (0): Conv1D(384, 256, kernel_size=[5], padding=2, data_format=NCL)\n",
      "          (1): ReLU()\n",
      "          (2): LayerNorm(normalized_shape=[256], epsilon=1e-05)\n",
      "          (3): Dropout(p=0.5, axis=None, mode=upscale_in_train)\n",
      "        )\n",
      "        (1): Sequential(\n",
      "          (0): Conv1D(256, 256, kernel_size=[5], padding=2, data_format=NCL)\n",
      "          (1): ReLU()\n",
      "          (2): LayerNorm(normalized_shape=[256], epsilon=1e-05)\n",
      "          (3): Dropout(p=0.5, axis=None, mode=upscale_in_train)\n",
      "        )\n",
      "        (2): Sequential(\n",
      "          (0): Conv1D(256, 256, kernel_size=[5], padding=2, data_format=NCL)\n",
      "          (1): ReLU()\n",
      "          (2): LayerNorm(normalized_shape=[256], epsilon=1e-05)\n",
      "          (3): Dropout(p=0.5, axis=None, mode=upscale_in_train)\n",
      "        )\n",
      "        (3): Sequential(\n",
      "          (0): Conv1D(256, 256, kernel_size=[5], padding=2, data_format=NCL)\n",
      "          (1): ReLU()\n",
      "          (2): LayerNorm(normalized_shape=[256], epsilon=1e-05)\n",
      "          (3): Dropout(p=0.5, axis=None, mode=upscale_in_train)\n",
      "        )\n",
      "        (4): Sequential(\n",
      "          (0): Conv1D(256, 256, kernel_size=[5], padding=2, data_format=NCL)\n",
      "          (1): ReLU()\n",
      "          (2): LayerNorm(normalized_shape=[256], epsilon=1e-05)\n",
      "          (3): Dropout(p=0.5, axis=None, mode=upscale_in_train)\n",
      "        )\n",
      "      )\n",
      "      (linear): Linear(in_features=256, out_features=1, dtype=float32)\n",
      "    )\n",
      "    (pitch_embed): Sequential(\n",
      "      (0): Conv1D(1, 384, kernel_size=[1], data_format=NCL)\n",
      "      (1): Dropout(p=0.0, axis=None, mode=upscale_in_train)\n",
      "    )\n",
      "    (energy_predictor): VariancePredictor(\n",
      "      (conv): LayerList(\n",
      "        (0): Sequential(\n",
      "          (0): Conv1D(384, 256, kernel_size=[3], padding=1, data_format=NCL)\n",
      "          (1): ReLU()\n",
      "          (2): LayerNorm(normalized_shape=[256], epsilon=1e-05)\n",
      "          (3): Dropout(p=0.5, axis=None, mode=upscale_in_train)\n",
      "        )\n",
      "        (1): Sequential(\n",
      "          (0): Conv1D(256, 256, kernel_size=[3], padding=1, data_format=NCL)\n",
      "          (1): ReLU()\n",
      "          (2): LayerNorm(normalized_shape=[256], epsilon=1e-05)\n",
      "          (3): Dropout(p=0.5, axis=None, mode=upscale_in_train)\n",
      "        )\n",
      "      )\n",
      "      (linear): Linear(in_features=256, out_features=1, dtype=float32)\n",
      "    )\n",
      "    (energy_embed): Sequential(\n",
      "      (0): Conv1D(1, 384, kernel_size=[1], data_format=NCL)\n",
      "      (1): Dropout(p=0.0, axis=None, mode=upscale_in_train)\n",
      "    )\n",
      "    (length_regulator): LengthRegulator()\n",
      "    (decoder): Encoder(\n",
      "      (embed): Sequential(\n",
      "        (0): ScaledPositionalEncoding(\n",
      "          (dropout): Dropout(p=0.2, axis=None, mode=upscale_in_train)\n",
      "        )\n",
      "      )\n",
      "      (encoders): MultiSequential(\n",
      "        (0): EncoderLayer(\n",
      "          (self_attn): MultiHeadedAttention(\n",
      "            (linear_q): Linear(in_features=384, out_features=384, dtype=float32)\n",
      "            (linear_k): Linear(in_features=384, out_features=384, dtype=float32)\n",
      "            (linear_v): Linear(in_features=384, out_features=384, dtype=float32)\n",
      "            (linear_out): Linear(in_features=384, out_features=384, dtype=float32)\n",
      "            (dropout): Dropout(p=0.2, axis=None, mode=upscale_in_train)\n",
      "          )\n",
      "          (feed_forward): MultiLayeredConv1d(\n",
      "            (w_1): Conv1D(384, 1536, kernel_size=[3], padding=1, data_format=NCL)\n",
      "            (w_2): Conv1D(1536, 384, kernel_size=[3], padding=1, data_format=NCL)\n",
      "            (dropout): Dropout(p=0.2, axis=None, mode=upscale_in_train)\n",
      "            (relu): ReLU()\n",
      "          )\n",
      "          (norm1): LayerNorm(normalized_shape=[384], epsilon=1e-05)\n",
      "          (norm2): LayerNorm(normalized_shape=[384], epsilon=1e-05)\n",
      "          (dropout): Dropout(p=0.2, axis=None, mode=upscale_in_train)\n",
      "        )\n",
      "        (1): EncoderLayer(\n",
      "          (self_attn): MultiHeadedAttention(\n",
      "            (linear_q): Linear(in_features=384, out_features=384, dtype=float32)\n",
      "            (linear_k): Linear(in_features=384, out_features=384, dtype=float32)\n",
      "            (linear_v): Linear(in_features=384, out_features=384, dtype=float32)\n",
      "            (linear_out): Linear(in_features=384, out_features=384, dtype=float32)\n",
      "            (dropout): Dropout(p=0.2, axis=None, mode=upscale_in_train)\n",
      "          )\n",
      "          (feed_forward): MultiLayeredConv1d(\n",
      "            (w_1): Conv1D(384, 1536, kernel_size=[3], padding=1, data_format=NCL)\n",
      "            (w_2): Conv1D(1536, 384, kernel_size=[3], padding=1, data_format=NCL)\n",
      "            (dropout): Dropout(p=0.2, axis=None, mode=upscale_in_train)\n",
      "            (relu): ReLU()\n",
      "          )\n",
      "          (norm1): LayerNorm(normalized_shape=[384], epsilon=1e-05)\n",
      "          (norm2): LayerNorm(normalized_shape=[384], epsilon=1e-05)\n",
      "          (dropout): Dropout(p=0.2, axis=None, mode=upscale_in_train)\n",
      "        )\n",
      "        (2): EncoderLayer(\n",
      "          (self_attn): MultiHeadedAttention(\n",
      "            (linear_q): Linear(in_features=384, out_features=384, dtype=float32)\n",
      "            (linear_k): Linear(in_features=384, out_features=384, dtype=float32)\n",
      "            (linear_v): Linear(in_features=384, out_features=384, dtype=float32)\n",
      "            (linear_out): Linear(in_features=384, out_features=384, dtype=float32)\n",
      "            (dropout): Dropout(p=0.2, axis=None, mode=upscale_in_train)\n",
      "          )\n",
      "          (feed_forward): MultiLayeredConv1d(\n",
      "            (w_1): Conv1D(384, 1536, kernel_size=[3], padding=1, data_format=NCL)\n",
      "            (w_2): Conv1D(1536, 384, kernel_size=[3], padding=1, data_format=NCL)\n",
      "            (dropout): Dropout(p=0.2, axis=None, mode=upscale_in_train)\n",
      "            (relu): ReLU()\n",
      "          )\n",
      "          (norm1): LayerNorm(normalized_shape=[384], epsilon=1e-05)\n",
      "          (norm2): LayerNorm(normalized_shape=[384], epsilon=1e-05)\n",
      "          (dropout): Dropout(p=0.2, axis=None, mode=upscale_in_train)\n",
      "        )\n",
      "        (3): EncoderLayer(\n",
      "          (self_attn): MultiHeadedAttention(\n",
      "            (linear_q): Linear(in_features=384, out_features=384, dtype=float32)\n",
      "            (linear_k): Linear(in_features=384, out_features=384, dtype=float32)\n",
      "            (linear_v): Linear(in_features=384, out_features=384, dtype=float32)\n",
      "            (linear_out): Linear(in_features=384, out_features=384, dtype=float32)\n",
      "            (dropout): Dropout(p=0.2, axis=None, mode=upscale_in_train)\n",
      "          )\n",
      "          (feed_forward): MultiLayeredConv1d(\n",
      "            (w_1): Conv1D(384, 1536, kernel_size=[3], padding=1, data_format=NCL)\n",
      "            (w_2): Conv1D(1536, 384, kernel_size=[3], padding=1, data_format=NCL)\n",
      "            (dropout): Dropout(p=0.2, axis=None, mode=upscale_in_train)\n",
      "            (relu): ReLU()\n",
      "          )\n",
      "          (norm1): LayerNorm(normalized_shape=[384], epsilon=1e-05)\n",
      "          (norm2): LayerNorm(normalized_shape=[384], epsilon=1e-05)\n",
      "          (dropout): Dropout(p=0.2, axis=None, mode=upscale_in_train)\n",
      "        )\n",
      "      )\n",
      "      (after_norm): LayerNorm(normalized_shape=[384], epsilon=1e-05)\n",
      "    )\n",
      "    (feat_out): Linear(in_features=384, out_features=80, dtype=float32)\n",
      "    (postnet): Postnet(\n",
      "      (postnet): LayerList(\n",
      "        (0): Sequential(\n",
      "          (0): Conv1D(80, 256, kernel_size=[5], padding=2, data_format=NCL)\n",
      "          (1): BatchNorm1D(num_features=256, momentum=0.9, epsilon=1e-05, data_format=NCL)\n",
      "          (2): Tanh()\n",
      "          (3): Dropout(p=0.5, axis=None, mode=upscale_in_train)\n",
      "        )\n",
      "        (1): Sequential(\n",
      "          (0): Conv1D(256, 256, kernel_size=[5], padding=2, data_format=NCL)\n",
      "          (1): BatchNorm1D(num_features=256, momentum=0.9, epsilon=1e-05, data_format=NCL)\n",
      "          (2): Tanh()\n",
      "          (3): Dropout(p=0.5, axis=None, mode=upscale_in_train)\n",
      "        )\n",
      "        (2): Sequential(\n",
      "          (0): Conv1D(256, 256, kernel_size=[5], padding=2, data_format=NCL)\n",
      "          (1): BatchNorm1D(num_features=256, momentum=0.9, epsilon=1e-05, data_format=NCL)\n",
      "          (2): Tanh()\n",
      "          (3): Dropout(p=0.5, axis=None, mode=upscale_in_train)\n",
      "        )\n",
      "        (3): Sequential(\n",
      "          (0): Conv1D(256, 256, kernel_size=[5], padding=2, data_format=NCL)\n",
      "          (1): BatchNorm1D(num_features=256, momentum=0.9, epsilon=1e-05, data_format=NCL)\n",
      "          (2): Tanh()\n",
      "          (3): Dropout(p=0.5, axis=None, mode=upscale_in_train)\n",
      "        )\n",
      "        (4): Sequential(\n",
      "          (0): Conv1D(256, 80, kernel_size=[5], padding=2, data_format=NCL)\n",
      "          (1): BatchNorm1D(num_features=80, momentum=0.9, epsilon=1e-05, data_format=NCL)\n",
      "          (2): Dropout(p=0.5, axis=None, mode=upscale_in_train)\n",
      "        )\n",
      "      )\n",
      "    )\n",
      "  )\n",
      ")\n",
      "FastSpeech2 done!\n"
     ]
    }
   ],
   "source": [
    "with open(phones_dict, \"r\") as f:\n",
    "    phn_id = [line.strip().split() for line in f.readlines()]\n",
    "vocab_size = len(phn_id)\n",
    "print(\"vocab_size:\", vocab_size)\n",
    "odim = fastspeech2_config.n_mels\n",
    "model = FastSpeech2(\n",
    "    idim=vocab_size, odim=odim, **fastspeech2_config[\"model\"])\n",
    "\n",
    "model.set_state_dict(paddle.load(fastspeech2_checkpoint)[\"main_params\"]) # 加载预训练模型参数\n",
    "model.eval() # 推理阶段不启用 batch norm 和 dropout\n",
    "stat = np.load(fastspeech2_stat)\n",
    "mu, std = stat # 读取数据预处理阶段数据集的均值和标准差\n",
    "mu, std = paddle.to_tensor(mu), paddle.to_tensor(std)\n",
    "fastspeech2_normalizer = ZScore(mu, std) # 构造归一化的新模型\n",
    "fastspeech2_inference = FastSpeech2Inference(fastspeech2_normalizer, model)\n",
    "fastspeech2_inference.eval()\n",
    "print(fastspeech2_inference)\n",
    "print(\"FastSpeech2 done!\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 调用声学模型"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "shepe of mel (n_frames x n_mels):\n",
      "[347, 80]\n"
     ]
    },
    {
     "data": {
      "image/png": "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
      "text/plain": [
       "<Figure size 648x432 with 2 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "with paddle.no_grad():\n",
    "    mel = fastspeech2_inference(phone_ids)\n",
    "print(\"shepe of mel (n_frames x n_mels):\")\n",
    "print(mel.shape)\n",
    "# 绘制声学模型输出的 mel 频谱\n",
    "fig, ax = plt.subplots(figsize=(9, 6))\n",
    "im = ax.imshow(mel.T, aspect='auto',origin='lower')\n",
    "fig.colorbar(im, ax=ax)\n",
    "plt.title('Mel Spectrogram')\n",
    "plt.xlabel('Time')\n",
    "plt.ylabel('Frequency')\n",
    "plt.tight_layout()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 声码器（Vocoder）\n",
    "<br></br>\n",
    "声码器将声学特征转换为波形。声码器需要解决的是 <font color=\"#ff0000\">“信息缺失的补全问题”</font>。信息缺失是指，在音频波形转换为频谱图的时候，存在**相位信息**的缺失，在频谱图转换为 mel 频谱图的时候，存在**频域压缩**导致的信息缺失；假设音频的采样率是16kHZ, 一帧的音频有 10ms，也就是说，1s 的音频有 16000 个采样点，而 1s 中包含 100 帧，每一帧有 160 个采样点，声码器的作用就是将一个频谱帧变成音频波形的 160 个采样点，所以声码器中一般会包含**上采样**模块。\n",
    "    \n",
    "<br></br>\n",
    "与声学模型类似，声码器也分为自回归模型和非自回归模型, 更细致的分类如下:\n",
    "\n",
    "- Autoregression\n",
    "    - WaveNet\n",
    "    - WaveRNN\n",
    "    - LPCNet\n",
    "- Flow\n",
    "    - WaveFlow\n",
    "    - WaveGlow\n",
    "    - FloWaveNet\n",
    "    - Parallel WaveNet\n",
    "- GAN\n",
    "    - WaveGAN\n",
    "    - Parallel WaveGAN\n",
    "    - MelGAN\n",
    "    - HiFi-GAN\n",
    "- VAE\n",
    "    - Wave-VAE\n",
    "- Diffusion\n",
    "    - WaveGrad\n",
    "    - DiffWave\n",
    "\n",
    "<br></br>\n",
    "PaddleSpeech TTS 主要实现了百度的 `WaveFlow` 和一些主流的 GAN Vocoder, 在本教程中，我们使用 `Parallel WaveGAN` 作为声码器。<\n",
    "\n",
    "<br></br> \n",
    "<img style=\"float: center;\" src=\"source/pwgan.png\" width=\"75%\"/> \n",
    "\n",
    "<br></br>\n",
    "各 GAN Vocoder 的生成器和判别器的 Loss 的区别如下表格所示:\n",
    "    \n",
    "Model  | Generator Loss |Discriminator Loss\n",
    ":-------------:| :------------:| :-----\n",
    "Parallel Wave GAN| adversial loss <br> Feature Matching  | Multi-Scale Discriminator |\n",
    "Mel GAN |adversial loss <br> Multi-resolution STFT loss  | adversial loss|\n",
    "Multi-Band Mel GAN | adversial loss <br> full band Multi-resolution STFT loss <br> sub band Multi-resolution STFT loss |Multi-Scale Discriminator|\n",
    "HiFi GAN |adversial loss <br> Feature Matching <br>  Mel-Spectrogram Loss | Multi-Scale Discriminator <br> Multi-Period Discriminator|\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 初始化声码器 Parallel WaveGAN"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Parallel WaveGAN done!\n"
     ]
    }
   ],
   "source": [
    "vocoder = PWGGenerator(**pwg_config[\"generator_params\"])\n",
    "\n",
    "vocoder.set_state_dict(paddle.load(pwg_checkpoint)[\"generator_params\"]) # 模型加载预训练参数\n",
    "vocoder.remove_weight_norm()\n",
    "vocoder.eval() # 推理阶段不启用 batch norm 和 dropout\n",
    "\n",
    "stat = np.load(pwg_stat) # 读取数据预处理阶段数据集的均值和标准差\n",
    "mu, std = stat\n",
    "mu, std = paddle.to_tensor(mu), paddle.to_tensor(std)\n",
    "pwg_normalizer = ZScore(mu, std)\n",
    "pwg_inference = PWGInference(pwg_normalizer, vocoder) # 构建归一化的模型\n",
    "pwg_inference.eval()\n",
    "print(\"Parallel WaveGAN done!\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 调用声码器"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "shepe of wav (time x n_channels):[104100, 1]\n"
     ]
    },
    {
     "data": {
      "image/png": "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
      "text/plain": [
       "<Figure size 648x432 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "with paddle.no_grad():\n",
    "    wav = pwg_inference(mel)\n",
    "print(\"shepe of wav (time x n_channels):%s\"%wav.shape)\n",
    "\n",
    "# 绘制声码器输出的波形图\n",
    "wave_data = wav.numpy().T\n",
    "time = np.arange(0, wave_data.shape[1]) * (1.0 / fastspeech2_config.fs)\n",
    "fig, ax = plt.subplots(figsize=(9, 6))\n",
    "plt.plot(time, wave_data[0])\n",
    "plt.title('Waveform')\n",
    "plt.xlabel('Time (seconds)')\n",
    "plt.ylabel('Amplitude (normed)')\n",
    "plt.tight_layout()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 播放音频"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "scrolled": true
   },
   "outputs": [],
   "source": [
    "dp.Audio(wav.numpy().T, rate=fastspeech2_config.fs)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 保存音频"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "scrolled": true
   },
   "outputs": [],
   "source": [
    "!mkdir output\n",
    "sf.write(\n",
    "    \"output/output.wav\",\n",
    "    wav.numpy(),\n",
    "    samplerate=fastspeech2_config.fs)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 进阶 —— 个性化调节\n",
    "<br></br>\n",
    "FastSpeech2 模型可以个性化地调节音素时长、音调和能量，通过一些简单的调节就可以获得一些有意思的效果。例如对于以下的原始音频`\"不要听信别人的谗言，我不是什么克隆人\"`。"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "原始音频\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "\n",
       "                <audio controls=\"controls\" >\n",
       "                    <source src=\"https://paddlespeech.bj.bcebos.com/Parakeet/docs/demos/speed/x1_001.wav\" type=\"audio/x-wav\" />\n",
       "                    Your browser does not support the audio element.\n",
       "                </audio>\n",
       "              "
      ],
      "text/plain": [
       "<IPython.lib.display.Audio object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "speed x 1.2\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "\n",
       "                <audio controls=\"controls\" >\n",
       "                    <source src=\"https://paddlespeech.bj.bcebos.com/Parakeet/docs/demos/speed/x1.2_001.wav\" type=\"audio/x-wav\" />\n",
       "                    Your browser does not support the audio element.\n",
       "                </audio>\n",
       "              "
      ],
      "text/plain": [
       "<IPython.lib.display.Audio object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "speed x 0.8\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "\n",
       "                <audio controls=\"controls\" >\n",
       "                    <source src=\"https://paddlespeech.bj.bcebos.com/Parakeet/docs/demos/speed/x0.8_001.wav\" type=\"audio/x-wav\" />\n",
       "                    Your browser does not support the audio element.\n",
       "                </audio>\n",
       "              "
      ],
      "text/plain": [
       "<IPython.lib.display.Audio object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "pitch x 1.3(童声)\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "\n",
       "                <audio controls=\"controls\" >\n",
       "                    <source src=\"https://paddlespeech.bj.bcebos.com/Parakeet/docs/demos/child_voice/001.wav\" type=\"audio/x-wav\" />\n",
       "                    Your browser does not support the audio element.\n",
       "                </audio>\n",
       "              "
      ],
      "text/plain": [
       "<IPython.lib.display.Audio object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "robot\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "\n",
       "                <audio controls=\"controls\" >\n",
       "                    <source src=\"https://paddlespeech.bj.bcebos.com/Parakeet/docs/demos/robot/001.wav\" type=\"audio/x-wav\" />\n",
       "                    Your browser does not support the audio element.\n",
       "                </audio>\n",
       "              "
      ],
      "text/plain": [
       "<IPython.lib.display.Audio object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "print(\"原始音频\")\n",
    "dp.display(dp.Audio(url=\"https://paddlespeech.bj.bcebos.com/Parakeet/docs/demos/speed/x1_001.wav\"))\n",
    "print(\"speed x 1.2\")\n",
    "dp.display(dp.Audio(url=\"https://paddlespeech.bj.bcebos.com/Parakeet/docs/demos/speed/x1.2_001.wav\"))\n",
    "print(\"speed x 0.8\")\n",
    "dp.display(dp.Audio(url=\"https://paddlespeech.bj.bcebos.com/Parakeet/docs/demos/speed/x0.8_001.wav\"))\n",
    "print(\"pitch x 1.3(童声)\")\n",
    "dp.display(dp.Audio(url=\"https://paddlespeech.bj.bcebos.com/Parakeet/docs/demos/child_voice/001.wav\"))\n",
    "print(\"robot\")\n",
    "dp.display(dp.Audio(url=\"https://paddlespeech.bj.bcebos.com/Parakeet/docs/demos/robot/001.wav\"))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "具体实现代码请参考 [Style Fs2](https://github.com/DeepSpeech/demos/style_fs2/run.sh)。"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<br></br>\n",
    "## 用 PaddleSpeech 训练 TTS 模型\n",
    "<br></br>\n",
    "PaddleSpeech 的 examples 是按照 数据集/模型 的结构安排的:\n",
    "```text\n",
    "examples   \n",
    "├── aishell3\n",
    "│   ├── README.md\n",
    "│   ├── tts3\n",
    "│   └── vc0\n",
    "├── csmsc\n",
    "│   ├── README.md\n",
    "│   ├── tts2\n",
    "│   ├── tts3\n",
    "│   ├── voc1\n",
    "│   └── voc3\n",
    "├── ...\n",
    "└── ...\n",
    "```\n",
    "我们在每个数据集的 README.md 介绍了子目录和模型的对应关系, 在 TTS 中有如下对应关系:\n",
    "```text\n",
    "tts0 - Tactron2\n",
    "tts1 - TransformerTTS\n",
    "tts2 - SpeedySpeech\n",
    "tts3 - FastSpeech2\n",
    "voc0 - WaveFlow\n",
    "voc1 - Parallel WaveGAN\n",
    "voc2 - MelGAN\n",
    "voc3 - MultiBand MelGAN\n",
    "```\n",
    "<br></br>\n",
    "### 基于 CSMCS 数据集训练 FastSpeech2 模型\n",
    "```bash\n",
    "git clone https://github.com/PaddlePaddle/PaddleSpeech.git\n",
    "cd examples/csmsc/tts\n",
    "```\n",
    "根据 README.md, 下载 CSMCS 数据集和其对应的强制对齐文件, 并放置在对应的位置\n",
    "```bash\n",
    "./run.sh\n",
    "```\n",
    "`run.sh` 中包含预处理、训练、合成、静态图推理等步骤:\n",
    "\n",
    "```bash\n",
    "#!/bin/bash\n",
    "set -e\n",
    "source path.sh\n",
    "gpus=0,1\n",
    "stage=0\n",
    "stop_stage=100\n",
    "conf_path=conf/default.yaml\n",
    "train_output_path=exp/default\n",
    "ckpt_name=snapshot_iter_153.pdz\n",
    "\n",
    "# with the following command, you can choice the stage range you want to run\n",
    "# such as `./run.sh --stage 0 --stop-stage 0`\n",
    "# this can not be mixed use with `$1`, `$2` ...\n",
    "source ${MAIN_ROOT}/utils/parse_options.sh || exit 1\n",
    "\n",
    "if [ ${stage} -le 0 ] && [ ${stop_stage} -ge 0 ]; then\n",
    "    # prepare data\n",
    "    bash ./local/preprocess.sh ${conf_path} || exit -1\n",
    "fi\n",
    "if [ ${stage} -le 1 ] && [ ${stop_stage} -ge 1 ]; then\n",
    "    # train model, all `ckpt` under `train_output_path/checkpoints/` dir\n",
    "    CUDA_VISIBLE_DEVICES=${gpus} ./local/train.sh ${conf_path} ${train_output_path} || exit -1\n",
    "fi\n",
    "if [ ${stage} -le 2 ] && [ ${stop_stage} -ge 2 ]; then\n",
    "    # synthesize, vocoder is pwgan\n",
    "    CUDA_VISIBLE_DEVICES=${gpus} ./local/synthesize.sh ${conf_path} ${train_output_path} ${ckpt_name} || exit -1\n",
    "fi\n",
    "if [ ${stage} -le 3 ] && [ ${stop_stage} -ge 3 ]; then\n",
    "    # synthesize_e2e, vocoder is pwgan\n",
    "    CUDA_VISIBLE_DEVICES=${gpus} ./local/synthesize_e2e.sh ${conf_path} ${train_output_path} ${ckpt_name} || exit -1\n",
    "fi\n",
    "if [ ${stage} -le 4 ] && [ ${stop_stage} -ge 4 ]; then\n",
    "    # inference with static model\n",
    "    CUDA_VISIBLE_DEVICES=${gpus} ./local/inference.sh ${train_output_path} || exit -1\n",
    "fi\n",
    "```\n",
    "\n",
    "### 基于 CSMCS 数据集训练 Parallel WaveGAN 模型\n",
    "```bash\n",
    "git clone https://github.com/PaddlePaddle/PaddleSpeech.git\n",
    "cd examples/csmsc/voc1\n",
    "```\n",
    "根据 README.md, 下载 CSMCS 数据集和其对应的强制对齐文件, 并放置在对应的位置\n",
    "```bash\n",
    "./run.sh\n",
    "```\n",
    "`run.sh` 中包含预处理、训练、合成等步骤:\n",
    "```bash\n",
    "#!/bin/bash\n",
    "set -e\n",
    "source path.sh\n",
    "gpus=0,1\n",
    "stage=0\n",
    "stop_stage=100\n",
    "conf_path=conf/default.yaml\n",
    "train_output_path=exp/default\n",
    "ckpt_name=snapshot_iter_5000.pdz\n",
    "\n",
    "# with the following command, you can choice the stage range you want to run\n",
    "# such as `./run.sh --stage 0 --stop-stage 0`\n",
    "# this can not be mixed use with `$1`, `$2` ...\n",
    "source ${MAIN_ROOT}/utils/parse_options.sh || exit 1\n",
    "\n",
    "if [ ${stage} -le 0 ] && [ ${stop_stage} -ge 0 ]; then\n",
    "    # prepare data\n",
    "    ./local/preprocess.sh ${conf_path} || exit -1\n",
    "fi\n",
    "if [ ${stage} -le 1 ] && [ ${stop_stage} -ge 1 ]; then\n",
    "    # train model, all `ckpt` under `train_output_path/checkpoints/` dir\n",
    "    CUDA_VISIBLE_DEVICES=${gpus} ./local/train.sh ${conf_path} ${train_output_path} || exit -1\n",
    "fi\n",
    "if [ ${stage} -le 2 ] && [ ${stop_stage} -ge 2 ]; then\n",
    "    # synthesize\n",
    "    CUDA_VISIBLE_DEVICES=${gpus} ./local/synthesize.sh ${conf_path} ${train_output_path} ${ckpt_name} || exit -1\n",
    "fi\n",
    "```"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# FAQ\n",
    "\n",
    "- 需要注意的问题\n",
    "- 经验与分享\n",
    "- 用户的其他问题"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# 作业\n",
    "在 CSMSC 数据集上利用 FastSpeech2 和 Parallel WaveGAN 实现一个中文 TTS 系统。"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# 关注 PaddleSpeech\n",
    "我们的 [Github地址](https://github.com/PaddlePaddle/PaddleSpeech/)，欢迎加入以下微信群参与讨论：\n",
    "    \n",
    "<img src=\"./source/wechat-group.png\" width=\"20%\" />"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3.7.0 64-bit ('yt_py37_develop': venv)",
   "language": "python",
   "name": "python37064bitytpy37developvenv88cd689abeac41d886f9210a708a170b"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.7.0"
  },
  "toc": {
   "base_numbering": 1,
   "nav_menu": {},
   "number_sections": true,
   "sideBar": true,
   "skip_h1_title": false,
   "title_cell": "Table of Contents",
   "title_sidebar": "Contents",
   "toc_cell": false,
   "toc_position": {
    "height": "607px",
    "left": "93px",
    "top": "66.1333px",
    "width": "200.594px"
   },
   "toc_section_display": true,
   "toc_window_display": true
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}