importing_and_feature_extraction.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Importing and Feature Extraction\n",
    "\n",
    "This Notebook handles the importing of audio files, the extraction of audio features and the writing of the dataset to a csv file."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# mathmatical functionaliy\n",
    "import numpy as np\n",
    "# operating system interaction, specifically querying for list of file names in a directory\n",
    "import os\n",
    "# audio feature extraction\n",
    "import librosa\n",
    "# write to a csv file\n",
    "import csv"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "write to csv complete!\n"
     ]
    }
   ],
   "source": [
    "## ======== FUNCTION DEFINITIONS ======== ##\n",
    "\n",
    "\n",
    "# returns attack length of synthesizer patch in seconds\n",
    "# y: sample data of the sound file as a list / sr: the sample rate\n",
    "def getAttack(y, sr):\n",
    "    maxVal = 0\n",
    "    maxSamp = 0\n",
    "    # searches for index of the sample with the largest value\n",
    "    for i in range(len(y)):\n",
    "        if(y[i] > maxVal):\n",
    "            maxVal = y[i]\n",
    "            maxSamp = i\n",
    "    # dividing this index by the sample rate returns the time between the start and this sample in seconds\n",
    "    attack = maxSamp / sr\n",
    "    return attack\n",
    "\n",
    "# returns integer corresponding to the class of sample. used to create target data for each example\n",
    "# name: name of the soundfile as string / classes: list of the titles of classes as present in soundfile names\n",
    "def getClass(name, classes):\n",
    "    # loops through each class name and tests whether it is present in file name\n",
    "    for i in range(len(classes)):\n",
    "        if classes[i] in name:\n",
    "            # if it is the the class index is set accordingly\n",
    "            c = classes[i]\n",
    "            break\n",
    "    return c\n",
    "\n",
    "# returns list of mean values for each MFCC\n",
    "# y: sample data of the sound file as a list / sr: the sample rate / n_mfcc: the number of MFCCs to calculate\n",
    "def getMeanMFCCs(y, sr, n_mfcc):\n",
    "    hop_length = 512\n",
    "    mfccs = librosa.feature.mfcc(y = y, sr = sr, hop_length = hop_length, n_mfcc = n_mfcc)\n",
    "    means = []\n",
    "    # loop through each mfcc and calculate its mean across the time period\n",
    "    for i in range(0, n_mfcc):\n",
    "        total = 0\n",
    "        for j in range(0, len(mfccs[0])):\n",
    "            total += mfccs[i][j]\n",
    "        mean = (total / np.shape(mfccs[0]))\n",
    "        means.append(mean[0])\n",
    "    return(means)\n",
    "\n",
    "# write a list of sublists to a csv file\n",
    "# data: the list of data\n",
    "def writeCSV(data, path, title):\n",
    "    myFiles = open(path + '/' + title, 'wb')\n",
    "    with myFiles:\n",
    "        writer = csv.writer(myFiles)\n",
    "        writer.writerows(data)\n",
    "    print('write to .CSV complete!')\n",
    "\n",
    "\n",
    "## ======== END FUNCTION DEFINITIONS ======== ##\n",
    "\n",
    "\n",
    "# path to directory to load files from. must only contain audio files or will throw error\n",
    "loadPath = '/Users/angelorussell/Desktop/pads'\n",
    "# returns a list of all names of files in a directory\n",
    "files = os.listdir(loadPath)\n",
    "\n",
    "# titles of each class as substrings to be found in file name strings\n",
    "classes = ['bass', 'pad', 'lead']\n",
    "\n",
    "# number of mel frequency cepstrum coeffcients to test for\n",
    "n_mfccs = 20\n",
    "\n",
    "# number of decimal places to round floating point values to\n",
    "round_n = 4\n",
    "\n",
    "# define list for data to be stored in csv\n",
    "data = [[]]\n",
    "# first sublist set to title of each collumn\n",
    "data[0].append('attack')\n",
    "for i in range(n_mfccs):\n",
    "    data[0].append('mfcc_'+str(i+1))\n",
    "data[0].append('class')\n",
    "\n",
    "# extract features and classes from each audio file, fill lists dat and tar respectively\n",
    "for i in range(len(files)):\n",
    "    # concatenate strings for the directory path and each file name to load audio files\n",
    "    concat = loadPath + '/' + files[i]\n",
    "    y, sr = librosa.load(concat)\n",
    "    # append new list and set first element to the attack rate of the sample\n",
    "    data.append([])\n",
    "    data[i+1].append(round(getAttack(y, sr), round_n))\n",
    "    # set the following n_mfcc elements to the means of each mfcc\n",
    "    #data[i+1].extend(round(getMeanMFCCs(y, sr, n_mfccs), round_n))\n",
    "    data[i+1].extend(round(elem, round_n) for elem in getMeanMFCCs(y, sr, n_mfccs))\n",
    "    # append \n",
    "    data[i+1].append(getClass(files[i], classes))\n",
    "\n",
    "# path to directory for csv to write to. must be different to the read path\n",
    "writePath = '/Users/angelorussell/Desktop'\n",
    "writeCSV(data, writePath, 'synthPatch.csv')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 2",
   "language": "python",
   "name": "python2"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 2
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython2",
   "version": "2.7.14"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}