Chapter_6_Notebook_Char_RNN_v2.ipynb

{
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Wepr3e5Rr7_E"
      },
      "source": [
        "This notebook provides the means to:\n",
        "\n",
        "- download a solo saxophone dataset from the repository provided.\n",
        "- preprocess the dataset into a text format suitable for training a Char-RNN model.\n",
        "- train a Char-RNN model.\n",
        "- generate notation from the trained model.\n",
        "\n",
        "All that is required is a Google Drive account and a HuggingFace account."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "xiJnF-NVsKIv"
      },
      "outputs": [],
      "source": [
        "# 1 CONNECT TO YOUR GOOGLE DRIVE\n",
        "\n",
        "import os, sys\n",
        "from google.colab import drive\n",
        "drive.mount('/content/mnt', force_remount=True)\n",
        "nb_path = '/content/notebooks'\n",
        "os.symlink('/content/mnt/My Drive/', nb_path)\n",
        "sys.path.insert(0, nb_path)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "fP9b-omVsfd6"
      },
      "outputs": [],
      "source": [
        "# 2 PULL CODEBASE FROM GITHUB\n",
        "\n",
        "%cd {nb_path}\n",
        "!git clone https://github.com/markhanslip/PhD_Ch6_Char-RNN\n",
        "%cd PhD_Ch6_Char-RNN"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "Hl4Icb_qsLjI"
      },
      "outputs": [],
      "source": [
        "# 3 INSTALL DEPENDENCIES (takes a while)\n",
        "\n",
        "!apt-get install lilypond\n",
        "!apt-get install poppler-utils\n",
        "\n",
        "!pip install praat-parselmouth\n",
        "!pip install pdf2image\n",
        "!pip install unidecode\n",
        "!pip install datasets"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "itMkg9Fvtx0l"
      },
      "outputs": [],
      "source": [
        "# 4 DOWNLOAD DATASET FROM HUGGINGFACE\n",
        "from datasets import load_dataset\n",
        "\n",
        "# PASTE YOUR HUGGING FACE ACCESS TOKEN HERE, GENERATE ONE FROM YOUR ACCOUNT AT https://huggingface.co/settings/tokens\n",
        "MY_ACCESS_TOKEN = \"\"\n",
        "\n",
        "dataset = load_dataset(\"markhanslip/markhanslip_phd_saxophone_data\", token = MY_ACCESS_TOKEN)\n",
        "\n",
        "audio = dataset['train']['audio'][0]['array']\n",
        "sr = dataset['train']['audio'][0]['sampling_rate']"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "uhy5RYG4t-zF"
      },
      "outputs": [],
      "source": [
        "# 5 CREATE TEXT DATA FROM AUDIO\n",
        "from DataProcessing import DataProcessing\n",
        "\n",
        "proc = DataProcessing(audio_array = audio, sample_rate = sr, output_file = \"ToneRows.txt\")\n",
        "proc.load_audio()\n",
        "proc.get_freqs()\n",
        "proc.freqs_to_MIDI()\n",
        "proc.get_onsets()\n",
        "proc.remove_zeros_for_pitches_only()\n",
        "# proc.augment_data()\n",
        "proc.pitches2file()"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "3JLpPJV-tX0A"
      },
      "outputs": [],
      "source": [
        "# 6 TRAIN Char-RNN MODEL\n",
        "from RNN import Trainer\n",
        "\n",
        "trainer = Trainer(training_data = \"ToneRows.txt\", epochs=500, layers=2)\n",
        "trainer.load_data()\n",
        "trainer.convert_to_tensor()\n",
        "trainer.build_model()\n",
        "trainer.train()\n",
        "trainer.save_model()"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "SavApFLovV6v"
      },
      "outputs": [],
      "source": [
        "from RNN import Generator\n",
        "import random\n",
        "\n",
        "generator = Generator(model_path = \"ToneRows.pth\")\n",
        "\n",
        "generator.load_model()\n",
        "\n",
        "for i in range(12):\n",
        "\n",
        "    generator.generate(prompt='{} '.format(str(random.randint(60, 76))), predict_len=int(48*1.5), temperature=0.95)\n",
        "\n",
        "    notes = generator.parse_output_pitch_only()\n",
        "    generator.pipe_to_lilypond_pitch_only(ly_file='ToneRows.ly')\n"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "KGQO_dkEPO31"
      },
      "outputs": [],
      "source": [
        "!lilypond ToneRows.ly"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "lXX-DJSuSL4k"
      },
      "outputs": [],
      "source": [
        "from pdf2image import convert_from_path\n",
        "\n",
        "images = convert_from_path(\"ToneRows.pdf\")\n",
        "images[0]"
      ]
    }
  ],
  "metadata": {
    "accelerator": "GPU",
    "colab": {
      "provenance": [],
      "gpuType": "T4"
    },
    "kernelspec": {
      "display_name": "Python 3",
      "name": "python3"
    },
    "language_info": {
      "name": "python"
    }
  },
  "nbformat": 4,
  "nbformat_minor": 0
}