TTS/dataset_analysis/CheckDatasetSNR.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "This notebook computes the average SNR a given Voice Dataset. If the SNR is too low, that might reduce the performance or prevent model to learn.\n",
    "\n",
    "To use this notebook, you need:\n",
    "- WADA SNR estimation: http://www.cs.cmu.edu/~robust/archive/algorithms/WADA_SNR_IS_2008/\n",
    "    1. extract in the same folder as this notebook\n",
    "    2. under MacOS you'll have to rebuild the executable. In the build folder: 1) remove existing .o files and 2) run make\n",
    "\n",
    "\n",
    "- FFMPEG: ```sudo apt-get install ffmpeg ```     \n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "import os, sys\n",
    "import glob\n",
    "import subprocess\n",
    "import tempfile\n",
    "import IPython\n",
    "import soundfile as sf\n",
    "import numpy as np\n",
    "from tqdm import tqdm\n",
    "from multiprocessing import Pool\n",
    "from matplotlib import pylab as plt\n",
    "%matplotlib inline"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# Set the meta parameters\n",
    "DATA_PATH = \"/home/erogol/Data/m-ai-labs/de_DE/by_book/female/eva_k/\"\n",
    "NUM_PROC = 1\n",
    "CURRENT_PATH = os.getcwd()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "def compute_file_snr(file_path):\n",
    "    \"\"\" Convert given file to required format with FFMPEG and process with WADA.\"\"\"\n",
    "    _, sr = sf.read(file_path)\n",
    "    new_file = file_path.replace(\".wav\", \"_tmp.wav\")\n",
    "    if sr != 16000:\n",
    "        command = f'ffmpeg -i \"{file_path}\" -ac 1 -acodec pcm_s16le -y -ar 16000 \"{new_file}\"'\n",
    "    else:\n",
    "        command = f'cp \"{file_path}\" \"{new_file}\"'\n",
    "    os.system(command)\n",
    "    command = [f'\"{CURRENT_PATH}/WadaSNR/Exe/WADASNR\"', f'-i \"{new_file}\"', f'-t \"{CURRENT_PATH}/WadaSNR/Exe/Alpha0.400000.txt\"', '-ifmt mswav']\n",
    "    output = subprocess.check_output(\" \".join(command), shell=True)\n",
    "    try:\n",
    "        output = float(output.split()[-3].decode(\"utf-8\"))\n",
    "    except:\n",
    "        raise RuntimeError(\" \".join(command))\n",
    "    os.system(f'rm \"{new_file}\"')\n",
    "    return output, file_path\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "wav_file = \"/home/erogol/Data/LJSpeech-1.1/wavs/LJ001-0001.wav\"\n",
    "output = compute_file_snr(wav_file)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "wav_files = glob.glob(f\"{DATA_PATH}/**/*.wav\", recursive=True)\n",
    "print(f\" > Number of wav files {len(wav_files)}\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "if NUM_PROC == 1:\n",
    "    file_snrs = [None] * len(wav_files) \n",
    "    for idx, wav_file in tqdm(enumerate(wav_files)):\n",
    "        tup = compute_file_snr(wav_file)\n",
    "        file_snrs[idx] = tup\n",
    "else:\n",
    "    with Pool(NUM_PROC) as pool:\n",
    "        file_snrs = list(tqdm(pool.imap(compute_file_snr, wav_files), total=len(wav_files)))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "snrs = [tup[0] for tup in file_snrs]\n",
    "\n",
    "error_idxs = np.where(np.isnan(snrs) == True)[0]\n",
    "error_files = [file_names[idx] for idx in error_idxs]\n",
    "\n",
    "file_snrs = [i for j, i in enumerate(file_snrs) if j not in error_idxs]\n",
    "file_names = [tup[1] for tup in file_snrs]\n",
    "snrs = [tup[0] for tup in file_snrs]\n",
    "file_idxs = np.argsort(snrs)\n",
    "\n",
    "\n",
    "print(f\" > Average SNR of the dataset:{np.mean(snrs)}\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "def output_snr_with_audio(idx):\n",
    "    file_idx = file_idxs[idx]\n",
    "    file_name = file_names[file_idx]\n",
    "    wav, sr = sf.read(file_name)\n",
    "    # multi channel to single channel\n",
    "    if len(wav.shape) == 2:\n",
    "        wav = wav[:, 0]\n",
    "    print(f\" > {file_name} - snr:{snrs[file_idx]}\")\n",
    "    IPython.display.display(IPython.display.Audio(wav, rate=sr))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# find worse SNR files\n",
    "N = 10  # number of files to fetch\n",
    "for i in range(N):\n",
    "    output_snr_with_audio(i)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# find best recordings\n",
    "N = 10  # number of files to fetch\n",
    "for i in range(N):\n",
    "    output_snr_with_audio(-i-1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "plt.hist(snrs, bins=100)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  }
 ],
 "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
}