04_demo_estimate_pitchclasses.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Demo: Predict pitch classes\n",
    "\n",
    "This notebook shows how to load pretrained models for all main experiments of the paper and to predict pitch classes for unseen audio recordings.\n",
    "\n",
    "&copy; Johannes Zeitler (johannes.zeitler@fau.de) and Christof Weiß (christof.weiss@audiolabs-erlangen.de), 2020/21"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import os\n",
    "import sys\n",
    "import numpy as np\n",
    "import librosa\n",
    "import matplotlib.pyplot as plt\n",
    "from harmonicCQT import HCQT, getHopSize\n",
    "import tensorflow as tf\n",
    "from FrameGenerators import PredictGenerator\n",
    "from LibFMP.C3 import smooth_downsample_feature_sequence, normalize_feature_sequence"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 1. Specify paramters, select and load model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "load and resample audio...\n",
      "compute hcqt...\n",
      "load model mCNN_WIDE_trainMusicNet-BeethovenPiano-WagnerPianoScore\n"
     ]
    }
   ],
   "source": [
    "fs = 22050 # audio sample rate\n",
    "rate = 10 # pitch class feature rate in Hz\n",
    "f = 'Schubert_D911-23_SC06'\n",
    "\n",
    "path_audio = os.path.join('Data', 'Schubert_Winterreise', 'audio_wav', f+'.wav')\n",
    "\n",
    "# load audio\n",
    "print('load and resample audio...')\n",
    "audioIn, _ = librosa.load(path_audio, sr=fs, mono=True)\n",
    "\n",
    "# Shorten audio to first 60 seconds\n",
    "audioIn[:60*fs]\n",
    "\n",
    "if rate not in (10, 50):\n",
    "    assert false, 'Error: output rate must be 10Hz or 50Hz'\n",
    "\n",
    "# HCQT config\n",
    "cqt_frameRate = 50 # desired cqt rate (actual cqt rate is approx. 57Hz due to power-of-two-constraints...)\n",
    "bottomNote = 'C1'\n",
    "bottomPitch = librosa.note_to_midi(bottomNote)\n",
    "numOctaves = 6\n",
    "numHarmonics = 5\n",
    "binsPerKey = 3\n",
    "hopSizeCQT = getHopSize(fs, cqt_frameRate, numOctaves, numHarmonics) # actual cqt hop size (corresponds to 57Hz)\n",
    "\n",
    "# compute HCQT\n",
    "print('compute hcqt...')\n",
    "hcqt, _, _ = HCQT(x=audioIn, fs=fs, frameRate=cqt_frameRate, f_min=bottomNote, binsPerKey=binsPerKey,\n",
    "                  numOctaves=numOctaves, numHarmonics=numHarmonics, centerNoteToBins=True,\n",
    "                  correctTuning=True)\n",
    "\n",
    "# tensorflow generator for feeding HCQT batches to the network. Decrease batch_size if insufficient memory.\n",
    "predictGen = PredictGenerator(hcqt, batch_size=100, numContextFrames=75, compression=10)\n",
    "\n",
    "\n",
    "#### Select model, load and compile the network ############################\n",
    "\n",
    "## Models from the cross-dataset experiments (Figure 3 of the paper): ##\n",
    "# modelName = 'mCNN_trainMusicNet' # Trained on MusicNet\n",
    "# modelName = 'mCNN_trainBeethovenPiano' ' # Trained on Beethoven Piano Sonatas\n",
    "# modelName = 'mCNN_trainWagnerPianoScore' # Trained on Wagner Ring\n",
    "# modelName = 'mCNN_trainMusicNet-BeethovenPiano-WagnerPianoScore' # Trained on Mix\n",
    "\n",
    "## Models from different architectures trained on a mixed dataset (Figure 4 of the paper): ##\n",
    "# modelName = 'mCNN_trainMusicNet-BeethovenPiano-WagnerPianoScore'  # Basic\n",
    "# modelName = 'mCNN_trainMusicNet-BeethovenPiano-WagnerPianoScore_LastLayer10' # BasicLast10\n",
    "# modelName = 'mCNN_WIDE_trainMusicNet-BeethovenPiano-WagnerPianoScore' # Wide\n",
    "# modelName = 'mCNN_WIDE-Inception_trainMusicNet-BeethovenPiano-WagnerPianoScore' # WideInception\n",
    "# modelName = 'mCNN_DEEP_trainMusicNet-BeethovenPiano-WagnerPianoScore' # Deep\n",
    "# modelName = 'mCNN_DEEP-ResNet_trainMusicNet-BeethovenPiano-WagnerPianoScore' # DeepResNet\n",
    "\n",
    "## Pitch-class features for chord recognition (Section 5 of the paper) ##\n",
    "modelName = 'mCNN_WIDE_trainMusicNet-BeethovenPiano-WagnerPianoScore'    # for chord reco on Schubert Winterreise\n",
    "# modelName = 'mCNN_WIDE_trainMusicNet-SMD-WagnerPianoScore'    # for chord reco on Beethoven Sonatas\n",
    "\n",
    "############################################################################\n",
    "\n",
    "print('load model ' + modelName)\n",
    "model = tf.keras.models.load_model(os.path.join('Models', modelName), compile=False)\n",
    "model.compile(loss = 'binary_crossentropy')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 2. Predict pitch classes"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "predict pitch classes...\n"
     ]
    }
   ],
   "source": [
    "print('predict pitch classes...')\n",
    "output = model.predict(predictGen).transpose()\n",
    "\n",
    "# interpolate chromas to be sampled at _exactly_ 50Hz (instead of 57Hz, see above)\n",
    "cqtTimes = np.arange(output.shape[1]) / (fs/hopSizeCQT)\n",
    "targetTimes = np.arange(0, np.max(cqtTimes), 1/cqt_frameRate)\n",
    "outputInterp = np.zeros((12, len(targetTimes)))\n",
    "for i in range(12):\n",
    "    outputInterp[i,:] = np.interp(targetTimes, cqtTimes, output[i,:])\n",
    "\n",
    "# if required, downsample to 10Hz\n",
    "if rate is 10:\n",
    "    outputInterp,_ = smooth_downsample_feature_sequence(outputInterp, fs, filt_len=5, down_sampling=5, w_type='boxcar')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 3. Plot predictions together with annotations (ground truth)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1076.4x216 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1076.4x216 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Plot predictions\n",
    "plt.figure(figsize=(14.95, 3))\n",
    "librosa.display.specshow(outputInterp, y_axis='chroma', x_axis='s', sr=fs, hop_length=fs/rate, cmap='gray_r')\n",
    "plt.xlim([0, 30])\n",
    "plt.title('Pitch class predictions')\n",
    "plt.show()\n",
    "\n",
    "# Load and plot annotations\n",
    "chromaPath_dest = os.path.join('Data', 'Schubert_Winterreise', 'Chroma')\n",
    "chroma = np.load(os.path.join(chromaPath_dest, f+'.npy'))\n",
    "\n",
    "plt.figure(figsize=(14.95, 3))\n",
    "librosa.display.specshow(chroma, y_axis='chroma', x_axis='s', sr=fs, hop_length=hopSizeCQT, cmap='gray_r')\n",
    "plt.xlim([0, 30])\n",
    "plt.title('Pitch class annotations')\n",
    "plt.show()"
   ]
  }
 ],
 "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": "3.7.9"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}