coqui-tts/speaker_encoder/notebooks/PlotUmapLibriTTS.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "checkpoint_10000.pth.tar  config.json\n",
      "checkpoint_20000.pth.tar  events.out.tfevents.1567518806.erogol-desktop\n",
      "checkpoint_30000.pth.tar\n"
     ]
    }
   ],
   "source": [
    "!ls /media/erogol/data_ssd/Models/libri_tts/speaker_encoder/libritts_360-September-03-2019_03+53PM-dc69074/"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      " > Setting up Audio Processor...\n",
      " | > sample_rate:16000\n",
      " | > num_mels:40\n",
      " | > min_level_db:-100\n",
      " | > frame_shift_ms:12.5\n",
      " | > frame_length_ms:50\n",
      " | > ref_level_db:20\n",
      " | > num_freq:1025\n",
      " | > power:None\n",
      " | > preemphasis:0.98\n",
      " | > griffin_lim_iters:None\n",
      " | > signal_norm:True\n",
      " | > symmetric_norm:True\n",
      " | > mel_fmin:0\n",
      " | > mel_fmax:8000.0\n",
      " | > max_norm:4.0\n",
      " | > clip_norm:True\n",
      " | > do_trim_silence:False\n",
      " | > n_fft:2048\n",
      " | > hop_length:200\n",
      " | > win_length:800\n"
     ]
    }
   ],
   "source": [
    "import torch\n",
    "import os\n",
    "import umap\n",
    "import random\n",
    "import glob\n",
    "import numpy as np\n",
    "\n",
    "from TTS.speaker_encoder.model import SpeakerEncoder\n",
    "from TTS.utils.audio import AudioProcessor\n",
    "from TTS.utils.generic_utils import load_config\n",
    "\n",
    "MODEL_PATH = \"/media/erogol/data_ssd/Models/libri_tts/speaker_encoder/libritts_360-half-September-28-2019_10+46AM-8565c50/best_model.pth.tar\"\n",
    "CONFIG_PATH = \"/media/erogol/data_ssd/Models/libri_tts/speaker_encoder/libritts_360-September-03-2019_03+53PM-dc69074/config.json\"\n",
    "EMBED_PATH = \"/home/erogol/Data/Libri-TTS/train-clean-360-embed_128/\"\n",
    "CONFIG = load_config(CONFIG_PATH)\n",
    "ap = AudioProcessor(**CONFIG['audio'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "116500\n"
     ]
    }
   ],
   "source": [
    "embed_files = glob.glob(EMBED_PATH+\"/**/*.npy\", recursive=True)\n",
    "print(len(embed_files))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'/home/erogol/Data/Libri-TTS/train-clean-360-embed_128/1025/75365/1025_75365_000002_000002.npy'"
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "embed_files[0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "904\n"
     ]
    }
   ],
   "source": [
    "speaker_paths = list(set([os.path.dirname(os.path.dirname(embed_file)) for embed_file in embed_files]))\n",
    "speaker_to_utter = {}\n",
    "for embed_file in embed_files:\n",
    "    speaker_path = os.path.dirname(os.path.dirname(embed_file))\n",
    "    try:\n",
    "        speaker_to_utter[speaker_path].append(embed_file)\n",
    "    except:\n",
    "        speaker_to_utter[speaker_path]=[embed_file]\n",
    "print(len(speaker_paths))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "embeds = []\n",
    "labels = []\n",
    "num_speakers = 20\n",
    "num_utters = 10\n",
    "speaker_idxs = np.random.choice(range(len(speaker_paths)), num_speakers, replace=False )\n",
    "\n",
    "for speaker_num, speaker_idx in enumerate(speaker_idxs):\n",
    "    speaker_path = speaker_paths[speaker_idx]\n",
    "    speakers_utter = speaker_to_utter[speaker_path]\n",
    "    utter_idxs = np.random.randint(0, len(speakers_utter) , num_utters)\n",
    "    for utter_idx in utter_idxs:\n",
    "            embed_path = speaker_to_utter[speaker_path][utter_idx]\n",
    "            embed = np.load(embed_path)\n",
    "            embeds.append(embed)\n",
    "            labels.append(speaker_num)\n",
    "embeds = np.concatenate(embeds)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "from matplotlib import cm\n",
    "viridis = cm.get_cmap('tab20', num_speakers)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/erogol/miniconda3/lib/python3.7/site-packages/sklearn/metrics/pairwise.py:258: RuntimeWarning: invalid value encountered in sqrt\n",
      "  return distances if squared else np.sqrt(distances, out=distances)\n",
      "/home/erogol/miniconda3/lib/python3.7/site-packages/umap/spectral.py:229: UserWarning: Embedding a total of 5 separate connected components using meta-embedding (experimental)\n",
      "  n_components\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1152x720 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "from matplotlib import pylab as plt\n",
    "%matplotlib inline\n",
    "\n",
    "model = umap.UMAP()\n",
    "projection = model.fit_transform(embeds)\n",
    "colors = [viridis(i) for i in labels]\n",
    "\n",
    "fig, ax = plt.subplots(figsize=(16, 10))\n",
    "im = ax.scatter(projection[:, 0], projection[:, 1], c=colors)\n",
    "plt.gca().set_aspect(\"equal\", \"datalim\")\n",
    "plt.title(\"UMAP projection\")\n",
    "plt.tight_layout()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "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": "3.7.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}