tothepoweroftom commited on
Commit
ee360dc
·
verified ·
1 Parent(s): 823689b

Update app.py

Browse files
Files changed (1) hide show
  1. app.py +54 -18
app.py CHANGED
@@ -7,17 +7,26 @@ import random
7
  from datasets import load_dataset
8
 
9
  # ==========================================
10
- # 1. DATA PREPARATION (MULTI-GENRE)
11
  # ==========================================
12
- print("Downloading and sorting dataset by genre... (This will take a minute on boot)")
13
- dataset = load_dataset("ailsntua/Chordonomicon", split="train", streaming=True)
14
 
15
- target_genres = ["pop", "rock", "jazz", "metal", "country", "blues", "r&b", "folk"]
 
 
 
 
 
 
 
 
 
16
  corpus_by_genre = {genre: set() for genre in target_genres}
17
  pattern = re.compile(r'<([^>]+)>\s*([^<]+)')
18
 
19
  for row in dataset:
20
- if all(len(progressions) >= 100 for progressions in corpus_by_genre.values()):
 
21
  break
22
 
23
  main_genre = str(row.get('main_genre', '')).lower()
@@ -26,7 +35,7 @@ for row in dataset:
26
 
27
  matched_genre = None
28
  for g in target_genres:
29
- if g in combined_genres and len(corpus_by_genre[g]) < 100:
30
  matched_genre = g
31
  break
32
 
@@ -42,10 +51,9 @@ for row in dataset:
42
 
43
  if chords and ('verse' in tag or 'chorus' in tag):
44
  corpus_by_genre[matched_genre].add(chords)
45
- if len(corpus_by_genre[matched_genre]) >= 100:
46
- break
47
 
48
  corpus_by_genre = {g: list(chords) for g, chords in corpus_by_genre.items()}
 
49
 
50
  # ==========================================
51
  # 2. MARKOV CHAIN LOGIC
@@ -110,6 +118,23 @@ NOTE_TO_MIDI = {'C': 60, 'Cs': 61, 'Db': 61, 'D': 62, 'Ds': 63, 'Eb': 63, 'E': 6
110
  MIDI_TO_NOTE = {60: 'C', 61: 'Db', 62: 'D', 63: 'Eb', 64: 'E', 65: 'F', 66: 'Gb', 67: 'G', 68: 'Ab', 69: 'A', 70: 'Bb', 71: 'B'}
111
  CHORD_INTERVALS = {'maj': [0, 4, 7], 'min': [0, 3, 7], '7': [0, 4, 7, 10], 'maj7': [0, 4, 7, 11], 'min7': [0, 3, 7, 10], 'sus4': [0, 5, 7], 'sus2': [0, 2, 7], 'dim': [0, 3, 6], 'no3d': [0, 7], '5': [0, 7]}
112
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
113
  def parse_chord_to_midi(chord_string):
114
  if not chord_string or chord_string == 'N': return [], ""
115
  chord_string = chord_string.split('/')[0]
@@ -139,12 +164,14 @@ def apply_voicing(pitches, voicing_type):
139
  elif voicing_type == "Open / Spread" and len(pitches) >= 3: pitches[0] -= 12; pitches[1] += 12
140
  return sorted(pitches) if voicing_type != "Open / Spread" else pitches
141
 
142
- def generate_audio_file(progression_string, instrument, transpose_semitones, voicing_type):
143
  if not progression_string.strip(): return None, None, ""
144
 
145
- is_metal = (instrument == "Metal Guitar")
146
- prog_num = 30 if is_metal else 0
147
- velocity = 110 if is_metal else 85
 
 
148
 
149
  midi = pretty_midi.PrettyMIDI(initial_tempo=120)
150
  inst = pretty_midi.Instrument(program=prog_num)
@@ -160,7 +187,10 @@ def generate_audio_file(progression_string, instrument, transpose_semitones, voi
160
  normalized_root = ((pitches[0] - 60) % 12) + 60
161
  transposed_chord_names.append(MIDI_TO_NOTE.get(normalized_root, "C") + quality)
162
 
163
- if is_metal: pitches = [p - 12 for p in pitches]
 
 
 
164
  pitches = apply_voicing(pitches, voicing_type)
165
 
166
  for pitch in pitches:
@@ -184,7 +214,7 @@ def app_logic(genre, order, length, instrument, transpose, voicing):
184
  corpus = corpus_by_genre.get(genre, [])
185
 
186
  if not corpus:
187
- return f"Error: No chords found for {genre}.", "", None, None
188
 
189
  model = train_markov_model(corpus, order=int(order))
190
  raw_chords = generate_progression(model, target_length=int(length), order=int(order))
@@ -201,11 +231,11 @@ with gr.Blocks(theme=gr.themes.Monochrome()) as demo:
201
 
202
  with gr.Row():
203
  with gr.Column(scale=1):
204
- gr.Markdown("### 1. Training Data")
205
  genre_dropdown = gr.Dropdown(
206
  choices=[g.capitalize() for g in target_genres],
207
  value="Pop",
208
- label="Dataset Genre (100 Progressions Each)"
209
  )
210
 
211
  gr.Markdown("### 2. Generation Settings")
@@ -216,10 +246,16 @@ with gr.Blocks(theme=gr.themes.Monochrome()) as demo:
216
  transpose_slider = gr.Slider(minimum=-12, maximum=12, step=1, value=0, label="Transpose (Semitones)")
217
  voicing_dropdown = gr.Dropdown(
218
  choices=["Root Position", "First Inversion", "Second Inversion", "Open / Spread", "Random Voice Leading"],
219
- value="Root Position",
220
  label="Chord Voicings"
221
  )
222
- instrument_dropdown = gr.Dropdown(choices=["Piano", "Metal Guitar"], value="Piano", label="Instrument")
 
 
 
 
 
 
223
 
224
  generate_btn = gr.Button("Generate Chords", variant="primary")
225
 
 
7
  from datasets import load_dataset
8
 
9
  # ==========================================
10
+ # 1. DATA PREPARATION (RANDOM SAMPLED)
11
  # ==========================================
12
+ MAX_PROGRESSIONS = 2000
 
13
 
14
+ print(f"Downloading and shuffling dataset... targeting {MAX_PROGRESSIONS} random progressions per genre.")
15
+
16
+ # The magic happens here: .shuffle(buffer_size=10000) mixes the stream on the fly!
17
+ dataset = load_dataset(
18
+ "ailsntua/Chordonomicon",
19
+ split="train",
20
+ streaming=True
21
+ ).shuffle(seed=random.randint(1, 1000), buffer_size=10000)
22
+
23
+ target_genres = ["pop", "rock", "jazz", "metal", "country", "blues", "r&b", "folk", "electronic"]
24
  corpus_by_genre = {genre: set() for genre in target_genres}
25
  pattern = re.compile(r'<([^>]+)>\s*([^<]+)')
26
 
27
  for row in dataset:
28
+ # Stop processing once EVERY genre has hit the max cap
29
+ if all(len(progressions) >= MAX_PROGRESSIONS for progressions in corpus_by_genre.values()):
30
  break
31
 
32
  main_genre = str(row.get('main_genre', '')).lower()
 
35
 
36
  matched_genre = None
37
  for g in target_genres:
38
+ if g in combined_genres and len(corpus_by_genre[g]) < MAX_PROGRESSIONS:
39
  matched_genre = g
40
  break
41
 
 
51
 
52
  if chords and ('verse' in tag or 'chorus' in tag):
53
  corpus_by_genre[matched_genre].add(chords)
 
 
54
 
55
  corpus_by_genre = {g: list(chords) for g, chords in corpus_by_genre.items()}
56
+ print("Randomized dataset loaded successfully!")
57
 
58
  # ==========================================
59
  # 2. MARKOV CHAIN LOGIC
 
118
  MIDI_TO_NOTE = {60: 'C', 61: 'Db', 62: 'D', 63: 'Eb', 64: 'E', 65: 'F', 66: 'Gb', 67: 'G', 68: 'Ab', 69: 'A', 70: 'Bb', 71: 'B'}
119
  CHORD_INTERVALS = {'maj': [0, 4, 7], 'min': [0, 3, 7], '7': [0, 4, 7, 10], 'maj7': [0, 4, 7, 11], 'min7': [0, 3, 7, 10], 'sus4': [0, 5, 7], 'sus2': [0, 2, 7], 'dim': [0, 3, 6], 'no3d': [0, 7], '5': [0, 7]}
120
 
121
+ # General MIDI Patch Numbers (0-indexed)
122
+ INSTRUMENT_MAP = {
123
+ "Acoustic Grand Piano": 0,
124
+ "Electric Piano (Rhodes)": 4,
125
+ "Drawbar Organ": 16,
126
+ "Acoustic Guitar (Nylon)": 24,
127
+ "Electric Guitar (Clean)": 27,
128
+ "Electric Guitar (Distortion)": 30,
129
+ "Synth Pad 1 (New Age)": 88,
130
+ "Synth Pad 2 (Warm)": 89,
131
+ "Synth Pad 3 (Polysynth)": 90,
132
+ "Synth Pad 4 (Choir)": 91,
133
+ "Synth Pad 7 (Halo)": 94,
134
+ "Synth Pad 8 (Sweep)": 95,
135
+ "Sci-Fi / Atmosphere": 103
136
+ }
137
+
138
  def parse_chord_to_midi(chord_string):
139
  if not chord_string or chord_string == 'N': return [], ""
140
  chord_string = chord_string.split('/')[0]
 
164
  elif voicing_type == "Open / Spread" and len(pitches) >= 3: pitches[0] -= 12; pitches[1] += 12
165
  return sorted(pitches) if voicing_type != "Open / Spread" else pitches
166
 
167
+ def generate_audio_file(progression_string, instrument_name, transpose_semitones, voicing_type):
168
  if not progression_string.strip(): return None, None, ""
169
 
170
+ # Look up the correct MIDI program number from our dictionary
171
+ prog_num = INSTRUMENT_MAP.get(instrument_name, 0)
172
+
173
+ # Give guitars and synths a slightly higher velocity so they cut through
174
+ velocity = 100 if prog_num > 20 else 85
175
 
176
  midi = pretty_midi.PrettyMIDI(initial_tempo=120)
177
  inst = pretty_midi.Instrument(program=prog_num)
 
187
  normalized_root = ((pitches[0] - 60) % 12) + 60
188
  transposed_chord_names.append(MIDI_TO_NOTE.get(normalized_root, "C") + quality)
189
 
190
+ # Drop the octave if it's a distorted metal guitar
191
+ if instrument_name == "Electric Guitar (Distortion)":
192
+ pitches = [p - 12 for p in pitches]
193
+
194
  pitches = apply_voicing(pitches, voicing_type)
195
 
196
  for pitch in pitches:
 
214
  corpus = corpus_by_genre.get(genre, [])
215
 
216
  if not corpus:
217
+ return f"Error: No chords found for {genre}. Wait for the dataset to finish loading in the console.", "", None, None
218
 
219
  model = train_markov_model(corpus, order=int(order))
220
  raw_chords = generate_progression(model, target_length=int(length), order=int(order))
 
231
 
232
  with gr.Row():
233
  with gr.Column(scale=1):
234
+ gr.Markdown(f"### 1. Training Data (Up to {MAX_PROGRESSIONS} songs per genre)")
235
  genre_dropdown = gr.Dropdown(
236
  choices=[g.capitalize() for g in target_genres],
237
  value="Pop",
238
+ label="Dataset Genre"
239
  )
240
 
241
  gr.Markdown("### 2. Generation Settings")
 
246
  transpose_slider = gr.Slider(minimum=-12, maximum=12, step=1, value=0, label="Transpose (Semitones)")
247
  voicing_dropdown = gr.Dropdown(
248
  choices=["Root Position", "First Inversion", "Second Inversion", "Open / Spread", "Random Voice Leading"],
249
+ value="Open / Spread", # Open spread sounds incredible on synth pads!
250
  label="Chord Voicings"
251
  )
252
+
253
+ # Feed the dictionary keys into the dropdown
254
+ instrument_dropdown = gr.Dropdown(
255
+ choices=list(INSTRUMENT_MAP.keys()),
256
+ value="Synth Pad 2 (Warm)",
257
+ label="Instrument"
258
+ )
259
 
260
  generate_btn = gr.Button("Generate Chords", variant="primary")
261