File size: 9,794 Bytes
c12a65c |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 |
{
"cells": [
{
"cell_type": "code",
"execution_count": 11,
"id": "61e10139",
"metadata": {},
"outputs": [],
"source": [
"import pickle\n",
"from music21 import *"
]
},
{
"cell_type": "code",
"execution_count": 3,
"id": "1a2b28be",
"metadata": {},
"outputs": [],
"source": [
"import torch\n",
"import torch.nn as nn\n",
"from torch.nn import functional as F\n",
"\n",
"device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')\n",
"\n",
"class GenerationRNN(nn.Module):\n",
" def __init__(self, input_size, hidden_size, output_size, n_layers=1):\n",
" super(GenerationRNN, self).__init__()\n",
" self.input_size = input_size\n",
" self.hidden_size = hidden_size\n",
" self.output_size = output_size\n",
" self.n_layers = n_layers\n",
" \n",
" self.embedding = nn.Embedding(input_size, hidden_size)\n",
" self.gru = nn.GRU(hidden_size, hidden_size, n_layers)\n",
" self.decoder = nn.Linear(hidden_size * n_layers, output_size)\n",
" \n",
" def forward(self, input, hidden):\n",
" # Creates embedding of the input texts\n",
" #print('initial input', input.size())\n",
" input = self.embedding(input.view(1, -1))\n",
" #print('input after embedding', input.size())\n",
" output, hidden = self.gru(input, hidden)\n",
" #print('output after gru', output.size())\n",
" #print('hidden after gru', hidden.size())\n",
" output = self.decoder(hidden.view(1, -1))\n",
" #print('output after decoder', output.size())\n",
" return output, hidden\n",
"\n",
" def init_hidden(self):\n",
" return torch.zeros(self.n_layers, 1, self.hidden_size).to(device)"
]
},
{
"cell_type": "code",
"execution_count": 4,
"id": "5b7120cf",
"metadata": {},
"outputs": [],
"source": [
"def predict_multimomial(net, prime_seq, predict_len, temperature=0.8):\n",
" '''\n",
" Arguments:\n",
" prime_seq - priming sequence (converted t)\n",
" predict_len - number of notes to predict for after prime sequence\n",
" '''\n",
" hidden = net.init_hidden()\n",
"\n",
" predicted = prime_seq.copy()\n",
" prime_seq = torch.tensor(prime_seq, dtype = torch.long).to(device)\n",
"\n",
"\n",
" # \"Building up\" the hidden state using the prime sequence\n",
" for p in range(len(prime_seq) - 1):\n",
" input = prime_seq[p]\n",
" _, hidden = net(input, hidden)\n",
" \n",
" # Last character of prime sequence\n",
" input = prime_seq[-1]\n",
" \n",
" # For every index to predict\n",
" for p in range(predict_len):\n",
"\n",
" # Pass the inputs to the model - output has dimension n_pitches - scores for each of the possible characters\n",
" output, hidden = net(input, hidden)\n",
" # Sample from the network output as a multinomial distribution\n",
" output = output.data.view(-1).div(temperature).exp()\n",
" predicted_id = torch.multinomial(output, 1)\n",
"\n",
" # Add predicted index to the list and use as next input\n",
" predicted.append(predicted_id.item()) \n",
" input = predicted_id\n",
"\n",
" return predicted"
]
},
{
"cell_type": "code",
"execution_count": 5,
"id": "8ce30142",
"metadata": {},
"outputs": [],
"source": [
"file_path = '/home/dmytro/ucu/music-generation/model.pkl'\n",
"with open(file_path, 'rb') as f:\n",
" model = pickle.load(f)"
]
},
{
"cell_type": "code",
"execution_count": 6,
"id": "84a2ea9b",
"metadata": {},
"outputs": [],
"source": [
"file_path = '/home/dmytro/ucu/music-generation/int_to_note.pkl'\n",
"with open(file_path, 'rb') as f:\n",
" int_to_note = pickle.load(f)"
]
},
{
"cell_type": "code",
"execution_count": 12,
"id": "07815507",
"metadata": {},
"outputs": [],
"source": [
"def create_midi(prediction_output):\n",
" \"\"\" convert the output from the prediction to notes and create a midi file\n",
" from the notes \"\"\"\n",
" offset = 0\n",
" output_notes = []\n",
"\n",
" # create note and chord objects based on the values generated by the model\n",
" for pattern in prediction_output:\n",
" # pattern is a chord\n",
" if ('.' in pattern) or pattern.isdigit():\n",
" notes_in_chord = pattern.split('.')\n",
" notes = []\n",
" for current_note in notes_in_chord:\n",
" new_note = note.Note(int(current_note))\n",
" new_note.storedInstrument = instrument.Piano()\n",
" notes.append(new_note)\n",
" new_chord = chord.Chord(notes)\n",
" new_chord.offset = offset\n",
" output_notes.append(new_chord)\n",
" # pattern is a note\n",
" else:\n",
" new_note = note.Note(pattern)\n",
" new_note.offset = offset\n",
" new_note.storedInstrument = instrument.Piano()\n",
" output_notes.append(new_note)\n",
"\n",
" # increase offset each iteration so that notes do not stack\n",
" offset += 0.5\n",
"\n",
" midi_stream = stream.Stream(output_notes)\n",
"\n",
" return midi_stream"
]
},
{
"cell_type": "code",
"execution_count": 30,
"id": "a70a41f1",
"metadata": {},
"outputs": [],
"source": [
"input_melody = [727,\n",
" 224,\n",
" 55,\n",
" 55,\n",
" 727,\n",
" 224,\n",
" 55]\n"
]
},
{
"cell_type": "code",
"execution_count": 28,
"id": "c9afc0c0",
"metadata": {},
"outputs": [],
"source": [
"generated_seq_multinomial = predict_multimomial(model, input_melody, predict_len = 100, temperature = 2.2)\n",
"generated_seq_multinomial = [int_to_note[e] for e in generated_seq_multinomial]\n",
"pred_midi_multinomial = create_midi(generated_seq_multinomial)"
]
},
{
"cell_type": "code",
"execution_count": 29,
"id": "99a1aabe",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"'/home/dmytro/ucu/music-generation/output/new_2.mid'"
]
},
"execution_count": 29,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"pred_midi_multinomial.write('midi', fp='result.mid')"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "ba84139a",
"metadata": {},
"outputs": [],
"source": [
"sound_font = \"/usr/share/sounds/sf2/FluidR3_GM.sf2\"\n",
"FluidSynth(sound_font).midi_to_audio('result.midi', 'result.wav')\n",
"return 'result.wav', 'result.midi'"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "0f4481b8",
"metadata": {},
"outputs": [],
"source": [
"def process_input():\n",
" pass"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "2f2e7a91",
"metadata": {},
"outputs": [],
"source": [
"midi_file_desc = \"\"\"Please entUpload your own MIDI file here (try to keep it small).\n",
"If you do not have a MIDI file, add some text and we will turn it into music!\n",
"\"\"\"\n",
"\n",
"article = \"\"\"# Pop Music Transformer\n",
"We are using a language model to create music by treating a musical standard MIDI a simple text, with tokens for note values, note duration, and separations to denote movement forward in time.\n",
"\n",
"This is all following the great work you can find [at this repo](https://github.com/bearpelican/musicautobot). Moreover check out [their full web app](http://musicautobot.com/). We use the pretrained model they created as well as the utilities for converting between MIDI, audio streams, numpy encodings, and WAV files.\n",
"\n",
"## Sonification\n",
"\n",
"This is the process of turning something not inherently musical into music. Here we do something pretty simple. We take your input text \"pretty cool\", get a sentiment score (hard coded right now, model TODO), and use a major progression if it's positive and a minor progression if it's negative, and then factor the score into the randomness of the generated music. We also take the text and extract a melody by taking any of the letters from A to G, which in the example is just \"E C\". With the simple \"E C\" melody and a major progression a musical idea is generated.\n",
"\"\"\"\n",
"\n",
"iface = gr.Interface(\n",
" fn=process_input, \n",
" inputs=[\n",
" gr.inputs.File(optional=True, label=midi_file_desc),\n",
" \"text\", \n",
" gr.inputs.Slider(0, 250, default=100, step=50),\n",
" gr.inputs.Radio([100, 200, 500], type=\"value\", default=100)\n",
" ], \n",
" outputs=[\"audio\", \"file\"],\n",
" article=article\n",
" # examples=['C major scale.midi']\n",
")\n",
"\n",
"iface.launch()"
]
}
],
"metadata": {
"kernelspec": {
"display_name": "Python 3 (ipykernel)",
"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": "3.9.12"
}
},
"nbformat": 4,
"nbformat_minor": 5
}
|