Merge branch 'master' of gitlab.idiap.ch:master-biometrics/04-lab-speaker

notebooks/features.01.fft.ipynb

@@ -14,7 +14,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 46,
+   "execution_count": 2,
    "metadata": {},
    "outputs": [],
    "source": [
@@ -38,7 +38,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 44,
+   "execution_count": 8,
    "metadata": {
     "scrolled": false
    },
@@ -104,7 +104,7 @@
     "                                                                 for x in fourier_cf])))\n",
     "\n",
     "# compute the Fourier using numpy's built in function\n",
-    "fourier_cf_numpy = numpy.fft.fft(signal)\n",
+    "fourier_cf_numpy = numpy.fft.fft(signal, n_samples)\n",
     "print('Fourier coeffficients with numpy implementation: \\n {0}'.format(fourier_cf_numpy))\n",
     "\n",
     "# plot the result of our Fourier transform\n",
@@ -121,8 +121,8 @@
     "# plot the result of numpy implementation of Fourier transform\n",
     "plt.subplot(1, 2, 2)\n",
     "plt.title('numpy implementation of Fourier transform')\n",
-    "plt.plot(freq, fourier_cf_numpy.real, label=\"real\")\n",
-    "plt.plot(freq, fourier_cf_numpy.imag, label=\"imag\")\n",
+    "plt.plot(range(0, n_samples), fourier_cf_numpy.real, label=\"real\")\n",
+    "plt.plot(range(0, n_samples), fourier_cf_numpy.imag, label=\"imag\")\n",
     "plt.legend(loc=1)\n",
     "\n",
     "plt.show()"

notebooks/features.02.spectrum.ipynb

 {
  "cells": [
   {
+   "attachments": {},
    "cell_type": "markdown",
    "metadata": {},
    "source": [
     "# From signal to spectrum/spectrogram\n",
     "\n",
-    "In this excerise, for a given audio signal, we compute spectrum and spectrogram step by step. We present all processing steps that are usually performed to compute a spectrogram. Here are the steps:\n",
+    "In this notebook, for a given audio signal, we compute spectrum and spectrogram step by step. The typical processing steps for computing spectrum (as part of computing MFCC features) are shown on this figure: \n",
+    "\n",
+    "![](figures/spectrum-diagram.png)\n",
+    "\n",
+    "Here are the steps in details:\n",
     "\n",
     "1. Read the audio signal from WAV file\n",
     "2. Compute the number of overlapping windows based on the desired window size and the overlapping length\n",

notebooks/features.03.mfcc.ipynb

@@ -10,8 +10,15 @@
     "\n",
     "The Mel-scale aims to mimic the non-linear human ear perception of sound, by being more discriminative at lower frequencies and less discriminative at higher frequencies. \n",
     "\n",
+    "In this notebook, starting from spectrum, we finish computing MFCC features by performing the steps illustrated in this figure:\n",
     "\n",
-    "\n"
+    "![](figures/mfcc-diagram.png)\n",
+    "\n",
+    "Here are the steps in more details:\n",
+    "\n",
+    "1. Build the Mel-scale filters (number of filters ``num_mel_filters`` is a parameter you can change) that would cover the length of the spectrum.\n",
+    "2. Multiply spectrum of each window by these Mel-scaled filters.\n",
+    "3. For each window, apply discrete cosine transform (DCT) on the filtered result and take the first ``num_mfcc`` values. These values are the MFCC features we were trying to compute."
    ]
   },
   {