TTS/notebooks/Benchmark.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "This is to test TTS models with benchmark sentences for speech synthesis.\n",
    "\n",
    "Before running this script please DON'T FORGET: \n",
    "- to set file paths.\n",
    "- to download related model files from TTS and WaveRNN.\n",
    "- to checkout right commit versions (given next to the model) of TTS and WaveRNN.\n",
    "- to set the right paths in the cell below.\n",
    "\n",
    "Repositories:\n",
    "- TTS: https://github.com/mozilla/TTS\n",
    "- WaveRNN: https://github.com/erogol/WaveRNN"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "TTS_PATH = \"/home/erogol/projects/\"\n",
    "WAVERNN_PATH =\"/home/erogol/projects/\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true,
    "scrolled": true
   },
   "outputs": [],
   "source": [
    "%load_ext autoreload\n",
    "%autoreload 2\n",
    "import os\n",
    "import sys\n",
    "import io\n",
    "import torch \n",
    "import time\n",
    "import numpy as np\n",
    "from collections import OrderedDict\n",
    "from matplotlib import pylab as plt\n",
    "\n",
    "%pylab inline\n",
    "rcParams[\"figure.figsize\"] = (16,5)\n",
    "\n",
    "# add libraries into environment\n",
    "sys.path.append(TTS_PATH) # set this if TTS is not installed globally\n",
    "sys.path.append(WAVERNN_PATH) # set this if TTS is not installed globally\n",
    "\n",
    "import librosa\n",
    "import librosa.display\n",
    "\n",
    "from TTS.models.tacotron import Tacotron \n",
    "from TTS.layers import *\n",
    "from TTS.utils.data import *\n",
    "from TTS.utils.audio import AudioProcessor\n",
    "from TTS.utils.generic_utils import load_config, setup_model\n",
    "from TTS.utils.text import text_to_sequence\n",
    "from TTS.utils.synthesis import synthesis\n",
    "from TTS.utils.visual import visualize\n",
    "\n",
    "import IPython\n",
    "from IPython.display import Audio\n",
    "\n",
    "import os\n",
    "os.environ['CUDA_VISIBLE_DEVICES']='1'\n",
    "os.environ['OMP_NUM_THREADS']='1'\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "def tts(model, text, CONFIG, use_cuda, ap, use_gl, speaker_id=None, figures=True):\n",
    "    t_1 = time.time()\n",
    "    waveform, alignment, mel_spec, mel_postnet_spec, stop_tokens = synthesis(model, text, CONFIG, use_cuda, ap, truncated=False, speaker_id=speaker_id, enable_eos_bos_chars=CONFIG.enable_eos_bos_chars)\n",
    "    if CONFIG.model == \"Tacotron\" and not use_gl:\n",
    "        mel_postnet_spec = ap.out_linear_to_mel(mel_postnet_spec.T).T\n",
    "    if not use_gl:\n",
    "        waveform = wavernn.generate(torch.FloatTensor(mel_postnet_spec.T).unsqueeze(0).cuda(), batched=batched_wavernn, target=11000, overlap=550)\n",
    "\n",
    "    print(\" >  Run-time: {}\".format(time.time() - t_1))\n",
    "    if figures:                                                                                                         \n",
    "        visualize(alignment, mel_postnet_spec, stop_tokens, text, ap.hop_length, CONFIG, mel_spec)                                                                       \n",
    "    IPython.display.display(Audio(waveform, rate=CONFIG.audio['sample_rate']))  \n",
    "    os.makedirs(OUT_FOLDER, exist_ok=True)\n",
    "    file_name = text.replace(\" \", \"_\").replace(\".\",\"\") + \".wav\"\n",
    "    out_path = os.path.join(OUT_FOLDER, file_name)\n",
    "    ap.save_wav(waveform, out_path)\n",
    "    return alignment, mel_postnet_spec, stop_tokens, waveform"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# Set constants\n",
    "ROOT_PATH = '/media/erogol/data_ssd/Data/models/mozilla_models/4845/'\n",
    "MODEL_PATH = ROOT_PATH + 'best_model.pth.tar'\n",
    "CONFIG_PATH = ROOT_PATH + '/config.json'\n",
    "OUT_FOLDER = \"/home/erogol/Dropbox/AudioSamples/benchmark_samples/\"\n",
    "CONFIG = load_config(CONFIG_PATH)\n",
    "VOCODER_MODEL_PATH = \"/media/erogol/data_ssd/Data/models/wavernn/mozilla/mozilla-May24-4763/model_checkpoints/best_model.pth.tar\"\n",
    "VOCODER_CONFIG_PATH = \"/media/erogol/data_ssd/Data/models/wavernn/mozilla/mozilla-May24-4763/config.json\"\n",
    "VOCODER_CONFIG = load_config(VOCODER_CONFIG_PATH)\n",
    "use_cuda = False\n",
    "\n",
    "# Set some config fields manually for testing\n",
    "# CONFIG.windowing = False\n",
    "# CONFIG.prenet_dropout = False\n",
    "# CONFIG.separate_stopnet = True\n",
    "# CONFIG.stopnet = True\n",
    "\n",
    "# Set the vocoder\n",
    "use_gl = True # use GL if True\n",
    "batched_wavernn = True    # use batched wavernn inference if True"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# LOAD TTS MODEL\n",
    "from utils.text.symbols import symbols, phonemes\n",
    "\n",
    "# load the model\n",
    "num_chars = len(phonemes) if CONFIG.use_phonemes else len(symbols)\n",
    "model = setup_model(num_chars, CONFIG)\n",
    "\n",
    "# load the audio processor\n",
    "ap = AudioProcessor(**CONFIG.audio)         \n",
    "\n",
    "\n",
    "# load model state\n",
    "if use_cuda:\n",
    "    cp = torch.load(MODEL_PATH)\n",
    "else:\n",
    "    cp = torch.load(MODEL_PATH, map_location=lambda storage, loc: storage)\n",
    "\n",
    "# load the model\n",
    "model.load_state_dict(cp['model'])\n",
    "if use_cuda:\n",
    "    model.cuda()\n",
    "model.eval()\n",
    "print(cp['step'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# LOAD WAVERNN\n",
    "if use_gl == False:\n",
    "    from WaveRNN.models.wavernn import Model\n",
    "    bits = 10\n",
    "\n",
    "    wavernn = Model(\n",
    "            rnn_dims=512,\n",
    "            fc_dims=512,\n",
    "            mode=\"mold\",\n",
    "            pad=2,\n",
    "            upsample_factors=VOCODER_CONFIG.upsample_factors,  # set this depending on dataset\n",
    "            feat_dims=VOCODER_CONFIG.audio[\"num_mels\"],\n",
    "            compute_dims=128,\n",
    "            res_out_dims=128,\n",
    "            res_blocks=10,\n",
    "            hop_length=ap.hop_length,\n",
    "            sample_rate=ap.sample_rate,\n",
    "        ).cuda()\n",
    "\n",
    "\n",
    "    check = torch.load(VOCODER_MODEL_PATH)\n",
    "    wavernn.load_state_dict(check['model'])\n",
    "    if use_cuda:\n",
    "        wavernn.cuda()\n",
    "    wavernn.eval();\n",
    "    print(check['step'])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Comparision with https://mycroft.ai/blog/available-voices/"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true,
    "scrolled": false
   },
   "outputs": [],
   "source": [
    "model.eval()\n",
    "model.decoder.max_decoder_steps = 2000\n",
    "speaker_id = 0\n",
    "sentence =  \"Bill got in the habit of asking himself “Is that thought true?” And if he wasn’t absolutely certain it was, he just let it go.\"\n",
    "align, spec, stop_tokens, wav = tts(model, sentence, CONFIG, use_cuda, ap, speaker_id=speaker_id, use_gl=use_gl, figures=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true,
    "scrolled": true
   },
   "outputs": [],
   "source": [
    "sentence = \"Be a voice, not an echo.\"  # 'echo' is not in training set. \n",
    "align, spec, stop_tokens, wav = tts(model, sentence, CONFIG, use_cuda, ap, speaker_id=speaker_id, use_gl=use_gl, figures=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "sentence = \"The human voice is the most perfect instrument of all.\"\n",
    "align, spec, stop_tokens, wav = tts(model, sentence, CONFIG, use_cuda, ap, speaker_id=speaker_id, use_gl=use_gl, figures=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "sentence = \"I'm sorry Dave. I'm afraid I can't do that.\"\n",
    "align, spec, stop_tokens, wav = tts(model, sentence, CONFIG, use_cuda, ap, speaker_id=speaker_id, use_gl=use_gl, figures=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true,
    "scrolled": true
   },
   "outputs": [],
   "source": [
    "sentence = \"This cake is great. It's so delicious and moist.\"\n",
    "align, spec, stop_tokens, wav = tts(model, sentence, CONFIG, use_cuda, ap, speaker_id=speaker_id, use_gl=use_gl, figures=True)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Comparison with https://keithito.github.io/audio-samples/"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "sentence = \"Generative adversarial network or variational auto-encoder.\"\n",
    "align, spec, stop_tokens, wav = tts(model, sentence, CONFIG, use_cuda, ap, speaker_id=speaker_id, use_gl=use_gl, figures=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "sentence = \"Scientists at the CERN laboratory say they have discovered a new particle.\"\n",
    "align, spec, stop_tokens, wav = tts(model, sentence, CONFIG, use_cuda, ap, speaker_id=speaker_id, use_gl=use_gl, figures=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "sentence = \"Here’s a way to measure the acute emotional intelligence that has never gone out of style.\"\n",
    "align, spec, stop_tokens, wav = tts(model, sentence, CONFIG, use_cuda, ap, speaker_id=speaker_id, use_gl=use_gl, figures=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "sentence = \"President Trump met with other leaders at the Group of 20 conference.\"\n",
    "align, spec, stop_tokens, wav = tts(model, sentence, CONFIG, use_cuda, ap, speaker_id=speaker_id, use_gl=use_gl, figures=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "sentence = \"The buses aren't the problem, they actually provide a solution.\"\n",
    "align, spec, stop_tokens, wav = tts(model, sentence, CONFIG, use_cuda, ap, speaker_id=speaker_id, use_gl=use_gl, figures=True)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Comparison with https://google.github.io/tacotron/publications/tacotron/index.html"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "sentence = \"Generative adversarial network or variational auto-encoder.\"\n",
    "align, spec, stop_tokens, wav = tts(model, sentence, CONFIG, use_cuda, ap, speaker_id=speaker_id, use_gl=use_gl, figures=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "sentence = \"Basilar membrane and otolaryngology are not auto-correlations.\"\n",
    "align, spec, stop_tokens, wav = tts(model, sentence, CONFIG, use_cuda, ap, speaker_id=speaker_id, use_gl=use_gl, figures=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "sentence = \" He has read the whole thing.\"\n",
    "align, spec, stop_tokens, wav = tts(model, sentence, CONFIG, use_cuda, ap, speaker_id=speaker_id, use_gl=use_gl, figures=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "sentence = \"He reads books.\"\n",
    "align, spec, stop_tokens, wav = tts(model, sentence, CONFIG, use_cuda, ap, speaker_id=speaker_id, use_gl=use_gl, figures=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true,
    "scrolled": false
   },
   "outputs": [],
   "source": [
    "sentence = \"Thisss isrealy awhsome.\"\n",
    "align, spec, stop_tokens, wav = tts(model, sentence, CONFIG, use_cuda, ap, speaker_id=speaker_id, use_gl=use_gl, figures=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true,
    "scrolled": false
   },
   "outputs": [],
   "source": [
    "sentence = \"This is your internet browser, Firefox.\"\n",
    "align, spec, stop_tokens, wav = tts(model, sentence, CONFIG, use_cuda, ap, speaker_id=speaker_id, use_gl=use_gl, figures=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "sentence = \"This is your internet browser Firefox.\"\n",
    "align, spec, stop_tokens, wav = tts(model, sentence, CONFIG, use_cuda, ap, speaker_id=speaker_id, use_gl=use_gl, figures=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "sentence = \"The quick brown fox jumps over the lazy dog.\"\n",
    "align, spec, stop_tokens, wav = tts(model, sentence, CONFIG, use_cuda, ap, speaker_id=speaker_id, use_gl=use_gl, figures=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "sentence = \"Does the quick brown fox jump over the lazy dog?\"\n",
    "align, spec, stop_tokens, wav = tts(model, sentence, CONFIG, use_cuda, ap, speaker_id=speaker_id, use_gl=use_gl, figures=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "sentence = \"Eren, how are you?\"\n",
    "align, spec, stop_tokens, wav = tts(model, sentence, CONFIG, use_cuda, ap, speaker_id=speaker_id, use_gl=use_gl, figures=True)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Hard Sentences"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "sentence = \"Encouraged, he started with a minute a day.\"\n",
    "align, spec, stop_tokens, wav = tts(model, sentence, CONFIG, use_cuda, ap, speaker_id=speaker_id, use_gl=use_gl, figures=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "sentence = \"His meditation consisted of “body scanning” which involved focusing his mind and energy on each section of the body from head to toe .\"\n",
    "align, spec, stop_tokens, wav = tts(model, sentence, CONFIG, use_cuda, ap, speaker_id=speaker_id, use_gl=use_gl, figures=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "sentence = \"Recent research at Harvard has shown meditating for as little as 8 weeks can actually increase the grey matter in the parts of the brain responsible for emotional regulation and learning . \"\n",
    "align, spec, stop_tokens, wav = tts(model, sentence, CONFIG, use_cuda, ap, speaker_id=speaker_id, use_gl=use_gl, figures=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "sentence = \"If he decided to watch TV he really watched it.\"\n",
    "align, spec, stop_tokens, wav = tts(model, sentence, CONFIG, use_cuda, ap, speaker_id=speaker_id, use_gl=use_gl, figures=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true,
    "scrolled": true
   },
   "outputs": [],
   "source": [
    "sentence = \"Often we try to bring about change through sheer effort and we put all of our energy into a new initiative .\"\n",
    "align, spec, stop_tokens, wav = tts(model, sentence, CONFIG, use_cuda, ap, speaker_id=speaker_id, use_gl=use_gl, figures=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# for twb dataset\n",
    "sentence = \"In our preparation for Easter, God in his providence offers us each year the season of Lent as a sacramental sign of our conversion.\"\n",
    "align, spec, stop_tokens, wav = tts(model, sentence, CONFIG, use_cuda, ap, speaker_id=speaker_id, use_gl=use_gl, figures=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# !zip benchmark_samples/samples.zip benchmark_samples/*"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3(mztts)",
   "language": "python",
   "name": "mztts"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.8"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}