{
"cells": [
{
"cell_type": "markdown",
"id": "7a9aa1cb",
"metadata": {},
"source": [
"## 미디 파일에 맞는 스타일 벡터 생성을 위한 곳"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "9c131428",
"metadata": {},
"outputs": [],
"source": [
"from music21 import note, chord\n",
"from music21.stream.base import Score\n",
"\n",
"def score_to_style_vector(score: Score) -> dict:\n",
" \"\"\"MIDI 파일(music21 score)을 분석하여 스타일 벡터를 생성하는 함수\"\"\"\n",
"\n",
" score.show('midi')\n",
"\n",
" # 1. Key & Mode\n",
" key_obj = score.analyze(\"key\")\n",
" key = key_obj.tonic.name # type: ignore\n",
" mode = key_obj.mode # type: ignore\n",
"\n",
" # 2. 멜로디 음표 리스트 추출\n",
" melody_notes = [n for n in score.flat.notes if isinstance(n, note.Note)]\n",
" pitches = [n.pitch.midi for n in melody_notes]\n",
" onsets = [n.offset for n in melody_notes]\n",
"\n",
" # 3. 코드 (화성) 분석\n",
" chords = [c for c in score.flat.getElementsByClass(chord.Chord)]\n",
" complex_chords = sum([1 for c in chords if len(c.pitches) > 3])\n",
" chord_complexity = complex_chords / (len(chords) + 1e-6) # 비율\n",
"\n",
" # 4. 멜로디 밀도 (단위 박자당 노트 수)\n",
" melody_density = len(melody_notes) / (score.highestTime + 1e-6) # type: ignore\n",
"\n",
" # 5. 싱코페이션 (off-beat 비율: 1/4 박자 단위 기준)\n",
" syncopation = sum([1 for o in onsets if (o % 1) != 0]) / (len(onsets) + 1e-6)\n",
"\n",
" # 6. 피치 범위\n",
" pitch_range = max(pitches) - min(pitches) if pitches else 0\n",
"\n",
" # 7. 사람이 생각하는 곡의 분위기\n",
" mood = input(\"분위기 입력 (이 중에서 선택: Happy, Chill, Emotional, Aggressive, Dreamy, Melodic): \")\n",
"\n",
" # 스타일 벡터 결과\n",
" style_vector = {\n",
" \"bpm\": 128, # 기본값 설정\n",
" \"key\": key,\n",
" \"mode\": mode,\n",
" \"chord_complexity\": round(chord_complexity, 3),\n",
" \"melody_density\": round(melody_density, 3),\n",
" \"syncopation\": round(syncopation, 3),\n",
" \"pitch_range\": pitch_range,\n",
" \"mood\": mood\n",
" }\n",
"\n",
" return style_vector"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "19ba84e1",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Skipping non-MIDI file: 0.mid\n",
"Skipping non-MIDI file: 1.mid\n",
"Skipping non-MIDI file: 10.mid\n",
"Skipping non-MIDI file: 11.mid\n",
"Skipping non-MIDI file: 12.mid\n",
"Skipping non-MIDI file: 13.mid\n",
"Skipping non-MIDI file: 14.mid\n",
"Skipping non-MIDI file: 15.mid\n",
"Skipping non-MIDI file: 16.mid\n",
"Skipping non-MIDI file: 17.mid\n",
"Skipping non-MIDI file: 18.mid\n",
"Skipping non-MIDI file: 19.mid\n",
"Skipping non-MIDI file: 2.mid\n",
"Skipping non-MIDI file: 20.mid\n",
"Skipping non-MIDI file: 21.mid\n",
"Skipping non-MIDI file: 22.mid\n",
"Skipping non-MIDI file: 23.mid\n",
"Skipping non-MIDI file: 24.mid\n",
"Skipping non-MIDI file: 25.mid\n",
"Skipping non-MIDI file: 26.mid\n",
"Skipping non-MIDI file: 27.mid\n",
"Skipping non-MIDI file: 28.mid\n",
"Skipping non-MIDI file: 29.mid\n",
"Skipping non-MIDI file: 3.mid\n",
"Skipping non-MIDI file: 30.mid\n",
"Skipping non-MIDI file: 31.mid\n",
"Skipping non-MIDI file: 32.mid\n",
"Skipping non-MIDI file: 33.mid\n",
"Skipping non-MIDI file: 4.mid\n",
"Skipping non-MIDI file: 5.mid\n",
"Skipping non-MIDI file: 6.mid\n",
"Skipping non-MIDI file: 7.mid\n",
"Skipping non-MIDI file: 8.mid\n",
"Skipping non-MIDI file: 9.mid\n"
]
}
],
"source": [
"from HarmonyMIDIToken import HarmonyMIDIToken as Tokenizer\n",
"import os\n",
"\n",
"#tokenized_data = []\n",
"\n",
"for filename in os.listdir(\"data\"):\n",
" if filename.endswith(\".mid\") and not filename in [i[\"name\"] for i in tokenized_data]:\n",
" MIDI = Tokenizer()\n",
" print(f\"file name: {filename}\")\n",
" MIDI.set_midi(os.path.join(\"data\", filename))\n",
" \n",
" vector = score_to_style_vector(MIDI.to_midi()) # type: ignore\n",
" \n",
" tokenized_data.append({\n",
" \"name\": filename,\n",
" \"vector\":vector,\n",
" \"token\":MIDI.token_id\n",
" })\n",
" else:\n",
" print(f\"Skipping non-MIDI or already Done file: {filename}\")\n",
" "
]
},
{
"cell_type": "code",
"execution_count": 2,
"id": "f79a6edf",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Tokenized MIDI data: 34\n"
]
}
],
"source": [
"print(\"Tokenized MIDI data:\", len(tokenized_data))"
]
},
{
"cell_type": "code",
"execution_count": 3,
"id": "21a9ba89",
"metadata": {},
"outputs": [],
"source": [
"import json # JSON 파일로 토크나이저 미디 데이터 저장\n",
"\n",
"with open('tokenized_midi_data.json', 'w') as f:\n",
" json.dump(tokenized_data, f, indent=4)"
]
},
{
"cell_type": "code",
"execution_count": 1,
"id": "dea532d4",
"metadata": {},
"outputs": [],
"source": [
"import json # JSON으로 저장한 토크나이저 미디 데이터 불러오기\n",
"\n",
"with open('tokenized_midi_data.json', 'r') as f:\n",
" tokenized_data = json.load(f)"
]
},
{
"cell_type": "markdown",
"id": "5a37da0b",
"metadata": {},
"source": [
"## 데이터 수정"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "995a6294",
"metadata": {},
"outputs": [
{
"data": {
"text/html": [
"\n",
" \n",
" \n",
" \n",
" "
],
"text/plain": [
""
]
},
"metadata": {},
"output_type": "display_data"
},
{
"data": {
"text/html": [
"\n",
" \n",
" \n",
" \n",
" "
],
"text/plain": [
""
]
},
"metadata": {},
"output_type": "display_data"
}
],
"source": [
"from HarmonyMIDIToken import HarmonyMIDIToken as Tokenizer\n",
"\n",
"for item in tokenized_data:\n",
" if item['vector']['mood'] == \"IDK\":\n",
" MIDI = Tokenizer()\n",
" \n",
" MIDI.set_id(item['token'])\n",
" MIDI.to_midi().show('midi')\n",
" item['vector']['mood'] = input(\"분위기 재 입력\")"
]
},
{
"cell_type": "code",
"execution_count": 4,
"id": "68c42254",
"metadata": {},
"outputs": [],
"source": [
"for item in tokenized_data:\n",
" if not \"name\" in item.keys():\n",
" print(item)\n",
" item['name'] = item['vector']\n",
" item.pop('vector')"
]
},
{
"cell_type": "code",
"execution_count": 2,
"id": "8cba97a6",
"metadata": {},
"outputs": [],
"source": [
"from HarmonyMIDIToken import HarmonyMIDIToken as Tokenizer\n",
"import os\n",
"\n",
"for item in tokenized_data:\n",
" filename = item['name']\n",
"\n",
" MIDI = Tokenizer()\n",
" MIDI.set_midi(os.path.join(\"data\", filename))\n",
"\n",
" try:\n",
" item['token'] = MIDI.token_id\n",
" except:\n",
" print(f\"{item['name']} 을 토크나이즈 하는데 실패함.\")"
]
},
{
"cell_type": "markdown",
"id": "df4c74b3",
"metadata": {},
"source": [
"## 데이터 전처리"
]
},
{
"cell_type": "code",
"execution_count": 7,
"id": "f7b77c0c",
"metadata": {},
"outputs": [],
"source": [
"from sklearn.preprocessing import OneHotEncoder, MinMaxScaler\n",
"from sklearn.compose import ColumnTransformer\n",
"from torch.nn.utils.rnn import pad_sequence\n",
"import torch\n",
"import pandas as pd\n",
"\n",
"vector_df = pd.DataFrame([item['vector'] for item in tokenized_data])\n",
"\n",
"# 전처리 파이프라인\n",
"preprocessor = ColumnTransformer([\n",
" (\"cat\", OneHotEncoder(sparse_output=False), [\"mode\", \"mood\", \"key\"]),\n",
" (\"num\", MinMaxScaler(), [\"bpm\", \"chord_complexity\", \"melody_density\", \"syncopation\", \"pitch_range\"])\n",
"])\n",
"\n",
"X = preprocessor.fit_transform(vector_df)\n",
"\n",
"#Tensor 변환\n",
"X_tensor = torch.tensor(X, dtype=torch.float32)\n",
"Y_tensor = []\n",
"for item in tokenized_data:\n",
" token = torch.tensor(item['token'], dtype=torch.long)\n",
" EOS = torch.tensor([[100, 15, 72, 14, 15, 58, 15]], dtype=torch.long)\n",
" Y_tensor.append(torch.cat([token, EOS], dim=0))\n",
"\n",
" #Y_tensor.append(token)\n",
"\n",
"# 패딩 처리\n",
"padded_Y = pad_sequence(Y_tensor, batch_first=True, padding_value=16) # (batch_size, max_len, 7)"
]
},
{
"cell_type": "code",
"execution_count": 8,
"id": "dd840788",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"X shape: torch.Size([34, 25])\n",
"Y shape: torch.Size([34, 128, 7])\n"
]
}
],
"source": [
"print(\"X shape:\", X_tensor.shape)\n",
"print(\"Y shape:\", padded_Y.shape)"
]
},
{
"cell_type": "code",
"execution_count": 9,
"id": "b4efc676",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Y example: tensor([[ 81, 3, 65, 1, 3, 53, 3],\n",
" [ 0, 1, 0, 1, 1, 0, 1],\n",
" [ 81, 2, 65, 1, 2, 53, 2],\n",
" [ 0, 1, 0, 1, 1, 0, 1],\n",
" [ 81, 1, 65, 1, 1, 53, 1],\n",
" [ 0, 1, 0, 1, 1, 0, 1],\n",
" [ 79, 2, 65, 1, 2, 53, 2],\n",
" [ 0, 1, 0, 1, 1, 0, 1],\n",
" [ 79, 2, 65, 1, 2, 53, 2],\n",
" [ 0, 1, 0, 1, 1, 0, 1],\n",
" [ 84, 1, 60, 8, 1, 55, 1],\n",
" [ 84, 2, 60, 8, 2, 55, 2],\n",
" [ 0, 2, 0, 1, 2, 0, 2],\n",
" [ 84, 2, 60, 8, 2, 55, 2],\n",
" [ 83, 1, 60, 8, 1, 55, 1],\n",
" [ 84, 2, 60, 8, 2, 55, 2],\n",
" [ 79, 1, 60, 8, 2, 55, 2],\n",
" [ 0, 1, 0, 1, 0, 0, 0],\n",
" [ 83, 2, 0, 1, 1, 0, 1],\n",
" [ 0, 0, 60, 8, 2, 55, 2],\n",
" [ 0, 2, 0, 1, 2, 48, 2],\n",
" [ 84, 3, 69, 5, 3, 57, 3],\n",
" [ 0, 1, 0, 1, 1, 0, 1],\n",
" [ 84, 2, 69, 5, 2, 57, 2],\n",
" [ 83, 1, 69, 5, 1, 57, 1],\n",
" [ 84, 2, 69, 5, 2, 57, 2],\n",
" [ 79, 1, 69, 5, 1, 57, 1],\n",
" [ 0, 1, 0, 1, 1, 0, 1],\n",
" [ 83, 2, 69, 5, 2, 57, 2],\n",
" [ 0, 2, 0, 1, 2, 50, 2],\n",
" [ 81, 2, 67, 1, 2, 55, 2],\n",
" [ 0, 1, 0, 1, 1, 0, 1],\n",
" [ 81, 2, 67, 1, 2, 55, 2],\n",
" [ 83, 1, 67, 1, 1, 55, 1],\n",
" [ 84, 2, 67, 1, 2, 55, 2],\n",
" [ 83, 2, 67, 1, 2, 55, 2],\n",
" [ 0, 1, 0, 1, 1, 0, 1],\n",
" [ 83, 2, 67, 1, 2, 55, 2],\n",
" [ 0, 2, 0, 1, 2, 48, 2],\n",
" [ 81, 3, 65, 1, 3, 53, 3],\n",
" [ 0, 1, 0, 1, 1, 0, 1],\n",
" [ 81, 2, 65, 1, 2, 53, 2],\n",
" [ 0, 1, 0, 1, 1, 0, 1],\n",
" [ 81, 1, 65, 1, 1, 53, 1],\n",
" [ 0, 1, 0, 1, 1, 0, 1],\n",
" [ 79, 2, 65, 1, 2, 53, 2],\n",
" [ 0, 1, 0, 1, 1, 0, 1],\n",
" [ 79, 2, 65, 1, 2, 53, 2],\n",
" [ 0, 1, 0, 1, 1, 0, 1],\n",
" [ 84, 1, 60, 8, 1, 55, 1],\n",
" [ 84, 2, 60, 8, 2, 55, 2],\n",
" [ 0, 2, 0, 1, 2, 0, 2],\n",
" [ 84, 2, 60, 8, 2, 55, 2],\n",
" [ 83, 1, 60, 8, 1, 55, 1],\n",
" [ 84, 2, 60, 8, 2, 55, 2],\n",
" [ 91, 1, 60, 8, 1, 55, 1],\n",
" [ 0, 1, 0, 1, 1, 0, 1],\n",
" [ 83, 2, 60, 8, 2, 55, 2],\n",
" [ 0, 2, 0, 1, 2, 48, 2],\n",
" [ 84, 3, 69, 5, 3, 57, 3],\n",
" [ 0, 1, 0, 1, 1, 0, 1],\n",
" [ 84, 2, 69, 5, 2, 57, 2],\n",
" [ 83, 1, 69, 5, 1, 57, 1],\n",
" [ 84, 2, 69, 5, 2, 57, 2],\n",
" [ 91, 1, 69, 5, 2, 57, 2],\n",
" [ 0, 1, 0, 1, 0, 0, 0],\n",
" [ 83, 2, 0, 1, 1, 0, 1],\n",
" [ 0, 2, 69, 5, 2, 57, 2],\n",
" [ 81, 2, 0, 1, 2, 50, 2],\n",
" [ 0, 1, 67, 1, 2, 55, 2],\n",
" [ 81, 2, 0, 1, 0, 0, 0],\n",
" [ 86, 1, 0, 1, 1, 0, 1],\n",
" [ 84, 2, 67, 1, 2, 55, 2],\n",
" [ 86, 2, 67, 1, 1, 55, 1],\n",
" [ 0, 0, 67, 1, 2, 55, 2],\n",
" [ 0, 1, 67, 1, 2, 55, 2],\n",
" [ 84, 2, 0, 1, 1, 0, 1],\n",
" [ 84, 2, 67, 1, 2, 55, 2],\n",
" [ 84, 2, 67, 1, 2, 48, 2],\n",
" [100, 15, 72, 14, 15, 58, 15],\n",
" [ 16, 16, 16, 16, 16, 16, 16],\n",
" [ 16, 16, 16, 16, 16, 16, 16],\n",
" [ 16, 16, 16, 16, 16, 16, 16],\n",
" [ 16, 16, 16, 16, 16, 16, 16],\n",
" [ 16, 16, 16, 16, 16, 16, 16],\n",
" [ 16, 16, 16, 16, 16, 16, 16],\n",
" [ 16, 16, 16, 16, 16, 16, 16],\n",
" [ 16, 16, 16, 16, 16, 16, 16],\n",
" [ 16, 16, 16, 16, 16, 16, 16],\n",
" [ 16, 16, 16, 16, 16, 16, 16],\n",
" [ 16, 16, 16, 16, 16, 16, 16],\n",
" [ 16, 16, 16, 16, 16, 16, 16],\n",
" [ 16, 16, 16, 16, 16, 16, 16],\n",
" [ 16, 16, 16, 16, 16, 16, 16],\n",
" [ 16, 16, 16, 16, 16, 16, 16],\n",
" [ 16, 16, 16, 16, 16, 16, 16],\n",
" [ 16, 16, 16, 16, 16, 16, 16],\n",
" [ 16, 16, 16, 16, 16, 16, 16],\n",
" [ 16, 16, 16, 16, 16, 16, 16],\n",
" [ 16, 16, 16, 16, 16, 16, 16],\n",
" [ 16, 16, 16, 16, 16, 16, 16],\n",
" [ 16, 16, 16, 16, 16, 16, 16],\n",
" [ 16, 16, 16, 16, 16, 16, 16],\n",
" [ 16, 16, 16, 16, 16, 16, 16],\n",
" [ 16, 16, 16, 16, 16, 16, 16],\n",
" [ 16, 16, 16, 16, 16, 16, 16],\n",
" [ 16, 16, 16, 16, 16, 16, 16],\n",
" [ 16, 16, 16, 16, 16, 16, 16],\n",
" [ 16, 16, 16, 16, 16, 16, 16],\n",
" [ 16, 16, 16, 16, 16, 16, 16],\n",
" [ 16, 16, 16, 16, 16, 16, 16],\n",
" [ 16, 16, 16, 16, 16, 16, 16],\n",
" [ 16, 16, 16, 16, 16, 16, 16],\n",
" [ 16, 16, 16, 16, 16, 16, 16],\n",
" [ 16, 16, 16, 16, 16, 16, 16],\n",
" [ 16, 16, 16, 16, 16, 16, 16],\n",
" [ 16, 16, 16, 16, 16, 16, 16],\n",
" [ 16, 16, 16, 16, 16, 16, 16],\n",
" [ 16, 16, 16, 16, 16, 16, 16],\n",
" [ 16, 16, 16, 16, 16, 16, 16],\n",
" [ 16, 16, 16, 16, 16, 16, 16],\n",
" [ 16, 16, 16, 16, 16, 16, 16],\n",
" [ 16, 16, 16, 16, 16, 16, 16],\n",
" [ 16, 16, 16, 16, 16, 16, 16],\n",
" [ 16, 16, 16, 16, 16, 16, 16],\n",
" [ 16, 16, 16, 16, 16, 16, 16],\n",
" [ 16, 16, 16, 16, 16, 16, 16],\n",
" [ 16, 16, 16, 16, 16, 16, 16]])\n"
]
}
],
"source": [
"print(\"Y example:\", padded_Y[0])"
]
},
{
"cell_type": "code",
"execution_count": 10,
"id": "4f5f5dc1",
"metadata": {},
"outputs": [],
"source": [
"import torch\n",
"\n",
"torch.save({\n",
" \"X\": X_tensor,\n",
" \"Y\": padded_Y\n",
"}, \"DIVA_dataset.pt\")"
]
},
{
"cell_type": "markdown",
"id": "224e349f",
"metadata": {},
"source": [
"## 전처리 끝!"
]
}
],
"metadata": {
"kernelspec": {
"display_name": "diva",
"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": "3.11.13"
}
},
"nbformat": 4,
"nbformat_minor": 5
}