{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {
    "Collapsed": "false"
   },
   "source": [
    "# Jupyter Notbook for phoneme coverage analysis\n",
    "\n",
    "This jupyter notebook checks dataset configured in config.json for phoneme coverage.\n",
    "As mentioned here https://github.com/mozilla/TTS/wiki/Dataset#what-makes-a-good-dataset a good phoneme coverage is recommended.\n",
    "\n",
    "Most parameters will be taken from config.json file in mozilla tts repo so please ensure it's configured correctly for your dataset.\n",
    "This notebook used lots of existring code from the TTS repo to ensure future compatibility.\n",
    "\n",
    "Many thanks to Neil Stoker supporting me on this topic :-).\n",
    "\n",
    "I provide this notebook without any warrenty but it's hopefully useful for your dataset analysis.\n",
    "\n",
    "Happy TTS'ing :-)\n",
    "\n",
    "Thorsten Müller\n",
    "\n",
    "* https://github.com/thorstenMueller/deep-learning-german-tts\n",
    "* https://discourse.mozilla.org/t/contributing-my-german-voice-for-tts/"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "Collapsed": "false"
   },
   "outputs": [],
   "source": [
    "# set some vars\n",
    "# TTS_PATH = \"/home/thorsten/___dev/tts/mozilla/TTS\"\n",
    "CONFIG_FILE = \"/path/to/config/config.json\"\n",
    "CHARS_TO_REMOVE = \".,:!?'\"\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "Collapsed": "false"
   },
   "outputs": [],
   "source": [
    "# import stuff\n",
    "from TTS.config import load_config\n",
    "from TTS.tts.datasets import load_tts_samples\n",
    "from TTS.tts.utils.text.tokenizer import TTSTokenizer\n",
    "from tqdm import tqdm\n",
    "from matplotlib import pylab as plt\n",
    "from multiprocessing import Pool, cpu_count\n",
    "\n",
    "# extra imports that might not be included in requirements.txt\n",
    "import collections\n",
    "import operator\n",
    "\n",
    "%matplotlib inline"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "Collapsed": "false",
    "tags": []
   },
   "outputs": [],
   "source": [
    "# Load config.json properties\n",
    "CONFIG = load_config(CONFIG_FILE)\n",
    "\n",
    "# Load some properties from config.json\n",
    "CONFIG_METADATA = load_tts_samples(CONFIG.datasets)[0]\n",
    "CONFIG_METADATA = CONFIG_METADATA\n",
    "CONFIG_DATASET = CONFIG.datasets[0]\n",
    "CONFIG_PHONEME_LANGUAGE = CONFIG.phoneme_language\n",
    "CONFIG_TEXT_CLEANER = CONFIG.text_cleaner\n",
    "CONFIG_ENABLE_EOS_BOS_CHARS = CONFIG.enable_eos_bos_chars\n",
    "\n",
    "# Will be printed on generated output graph\n",
    "CONFIG_RUN_NAME = CONFIG.run_name\n",
    "CONFIG_RUN_DESC = CONFIG.run_description\n",
    "\n",
    "# Needed to convert text to phonemes and phonemes to ids\n",
    "tokenizer, config = TTSTokenizer.init_from_config(CONFIG)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "Collapsed": "false",
    "tags": []
   },
   "outputs": [],
   "source": [
    "# print some debug information on loaded config values\n",
    "print(\" > Run name: \" + CONFIG_RUN_NAME + \" (\" + CONFIG_RUN_DESC + \")\")\n",
    "print(\" > Dataset files: \" + str(len(CONFIG_METADATA)))\n",
    "print(\" > Phoneme language: \" + CONFIG_PHONEME_LANGUAGE)\n",
    "print(\" > Used text cleaner: \" + CONFIG_TEXT_CLEANER)\n",
    "print(\" > Enable eos bos chars: \" + str(CONFIG_ENABLE_EOS_BOS_CHARS))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "def get_phoneme_from_sequence(text):\n",
    "    temp_list = []\n",
    "    if len(text[\"text\"]) > 0:\n",
    "        #temp_text = text[0].rstrip('\\n')\n",
    "        temp_text = text[\"text\"].rstrip('\\n')\n",
    "        for rm_bad_chars in CHARS_TO_REMOVE:\n",
    "            temp_text = temp_text.replace(rm_bad_chars,\"\")\n",
    "        seq = tokenizer.text_to_ids(temp_text)\n",
    "        text = tokenizer.ids_to_text(seq)\n",
    "        text = text.replace(\" \",\"\")\n",
    "        temp_list.append(text)\n",
    "    return temp_list"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "Collapsed": "false",
    "tags": []
   },
   "outputs": [],
   "source": [
    "# Get phonemes from metadata\n",
    "phonemes = []\n",
    "\n",
    "with Pool(cpu_count()-1) as p:\n",
    "    \n",
    "    phonemes = list(tqdm(p.imap(get_phoneme_from_sequence, CONFIG_METADATA), total=len(CONFIG_METADATA)))\n",
    "    phonemes = [i for sub in phonemes for i in sub]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "Collapsed": "false",
    "tags": []
   },
   "outputs": [],
   "source": [
    "s = \"\"\n",
    "phonemeString = s.join(phonemes)\n",
    "\n",
    "d = {}\n",
    "collections._count_elements(d, phonemeString)\n",
    "sorted_d = dict(sorted(d.items(), key=operator.itemgetter(1),reverse=True))\n",
    "\n",
    "# remove useless keys\n",
    "sorted_d.pop(' ', None)\n",
    "sorted_d.pop('ˈ', None)\n",
    "\n",
    "phonemesSum = len(phonemeString.replace(\" \",\"\"))\n",
    "\n",
    "print(\"Dataset contains \" + str(len(sorted_d)) + \" different ipa phonemes.\")\n",
    "print(\"Dataset consists of \" + str(phonemesSum) + \" phonemes\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "Collapsed": "false",
    "tags": []
   },
   "outputs": [],
   "source": [
    "print(\"5 rarest phonemes\")\n",
    "\n",
    "rareList = dict(sorted(sorted_d.items(), key=operator.itemgetter(1), reverse=False)[:5])\n",
    "for key, value in rareList.items():\n",
    "    print(key + \" --> \" + str(value) + \" occurrences\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "Collapsed": "false"
   },
   "outputs": [],
   "source": [
    "# create plot from analysis result\n",
    "\n",
    "x = []\n",
    "y = []\n",
    "\n",
    "for key, value in sorted_d.items():\n",
    "    x.append(key)\n",
    "    y.append(value)\n",
    "\n",
    "plt.figure(figsize=(50,50))\n",
    "plt.title(\"Phoneme coverage for \" + CONFIG_RUN_NAME + \" (\" + CONFIG_RUN_DESC + \")\", fontsize=50)\n",
    "plt.xticks(fontsize=50)\n",
    "plt.yticks(fontsize=50)\n",
    "plt.barh(x,y, align='center', alpha=1.0)\n",
    "plt.gca().invert_yaxis()\n",
    "plt.ylabel('phoneme', fontsize=50)\n",
    "plt.xlabel('occurrences', fontsize=50)\n",
    "\n",
    "for i, v in enumerate(y):\n",
    "    plt.text(v + 2, i - .2, str(v), fontsize=20)\n",
    "    plt.text(v + 2, i + .2, \"(\" + str(round(100/phonemesSum * v,2)) + \"%)\", fontsize=20)\n",
    "    \n",
    "    \n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "Collapsed": "false"
   },
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.9.12"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}