File size: 11,297 Bytes
9c7e38e
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
"""
Data preprocessing for grid-based multi-instrument music generation.

- Loads MIDI files from a directory (e.g., MusicNet).
- Maps raw MIDI instruments to 4 families: Strings, Woodwinds, Brass, Percussion.
- Quantizes notes to a fixed grid at 120 BPM, with 8 steps per bar.
- Limits melodic pitches to one octave (C4–B4) and maps to indices 0–11.
- Maps percussion to Kick/Snare/Hi-hat rows.
- Builds sequences of length SEQ_LEN (in steps), with each time step a 39-dim vector:
    [12 strings, 12 woodwinds, 12 brass, 3 perc]
- Saves X (inputs) and Y (targets) arrays to disk.
"""

from typing import List, Tuple, Dict

import glob
import os

import numpy as np
import pretty_midi

import config


# One octave: C4 (60) to B4 (71)
MELODIC_BASE_PITCH: int = 60
MELODIC_MAX_PITCH: int = MELODIC_BASE_PITCH + config.PITCHES_MELODIC - 1  # 71

# Percussion pitches
PERC_KICK_PITCH: int = 36   # Bass Drum 1
PERC_SNARE_PITCH: int = 38  # Acoustic Snare
PERC_HIHAT_PITCH: int = 42  # Closed Hi-Hat


def load_midi_paths_from_dir(midi_root: str, pattern: str = "**/*.mid*") -> List[str]:
    """
    Recursively collect MIDI file paths from a root directory.

    Args:
        midi_root: Root directory containing MIDI files.
        pattern: Glob pattern relative to root.

    Returns:
        List of file paths.
    """
    search_path = os.path.join(midi_root, pattern)
    filepaths = glob.glob(search_path, recursive=True)
    return filepaths


# ============
# Instrument family mapping
# ============

def get_family_for_program(program: int, is_drum: bool) -> str:
    """
    Map a MIDI program number to one of the four families.

    Args:
        program: MIDI program number (0–127).
        is_drum: Whether this is a drum channel.

    Returns:
        One of: "strings", "woodwinds", "brass", "perc".
    """
    if is_drum:
        return "perc"

    # Very simple heuristic based on General MIDI program ranges.
    # You can refine this mapping later as needed.

    # Strings: Violins, Violas, Celli, etc. (40–47)
    if 40 <= program <= 47:
        return "strings"

    # Woodwinds/Flutes/Clarinets: Flute (73), Clarinet (71), etc.
    if 64 <= program <= 79:
        return "woodwinds"

    # Brass: Trumpet, Trombone, etc. (56–63)
    if 56 <= program <= 63:
        return "brass"

    # Fallback: treat others as strings
    return "strings"


# ============
# Quantization and grid building
# ============

def time_to_step_index(t: float) -> int:
    """
    Convert time (seconds) to global step index on the quantized grid.

    Using config.SECONDS_PER_STEP.

    Args:
        t: Time in seconds.

    Returns:
        Global step index (int).
    """
    return int(np.floor(t / config.SECONDS_PER_STEP))


def clip_pitch_to_melodic_range(pitch: int) -> int:
    """
    Map MIDI pitch to [MELODIC_BASE_PITCH, MELODIC_MAX_PITCH] if possible.

    Returns:
        -1 if outside range (should be ignored), or
        pitch_idx in [0, PITCHES_MELODIC-1] otherwise.
    """
    if pitch < MELODIC_BASE_PITCH or pitch > MELODIC_MAX_PITCH:
        return -1
    return pitch - MELODIC_BASE_PITCH


def instrument_to_family_grids(
    pm: pretty_midi.PrettyMIDI,
    max_bars: int = None,
) -> Dict[str, np.ndarray]:
    """
    Convert a PrettyMIDI object into per-family grids.

    Grids have shape:
        strings:   [bars, STEPS_PER_BAR, 12]
        woodwinds: [bars, STEPS_PER_BAR, 12]
        brass:     [bars, STEPS_PER_BAR, 12]
        perc:      [bars, STEPS_PER_BAR, 3]

    Args:
        pm: PrettyMIDI object.
        max_bars: Optional cap on number of bars.

    Returns:
        Dict of family → grid.
    """
    total_seconds = pm.get_end_time()
    bar_duration = config.SECONDS_PER_BAR
    total_bars = int(np.ceil(total_seconds / bar_duration))

    if max_bars is not None:
        total_bars = min(total_bars, max_bars)

    T = total_bars * config.STEPS_PER_BAR

    strings = np.zeros((total_bars, config.STEPS_PER_BAR, config.PITCHES_MELODIC), dtype=np.float32)
    woodwinds = np.zeros_like(strings)
    brass = np.zeros_like(strings)
    perc = np.zeros((total_bars, config.STEPS_PER_BAR, config.PITCHES_PERC), dtype=np.float32)

    # Helper to convert time to (bar_idx, step_idx)
    def t_to_bar_step(t: float) -> Tuple[int, int]:
        step_idx_global = time_to_step_index(t)
        bar_idx = step_idx_global // config.STEPS_PER_BAR
        step_idx = step_idx_global % config.STEPS_PER_BAR
        if bar_idx >= total_bars:
            bar_idx = total_bars - 1
        return bar_idx, step_idx

    for inst in pm.instruments:
        is_drum = inst.is_drum
        family = get_family_for_program(inst.program, is_drum=is_drum)

        if family == "perc":
            for note in inst.notes:
                b, s = t_to_bar_step(note.start)
                if note.pitch == PERC_KICK_PITCH:
                    k = 0
                elif note.pitch == PERC_SNARE_PITCH:
                    k = 1
                elif note.pitch == PERC_HIHAT_PITCH:
                    k = 2
                else:
                    continue
                perc[b, s, k] = 1.0
        else:
            # Select grid
            if family == "strings":
                grid = strings
            elif family == "woodwinds":
                grid = woodwinds
            elif family == "brass":
                grid = brass
            else:
                grid = strings  # fallback

            for note in inst.notes:
                b, s = t_to_bar_step(note.start)
                pitch_idx = clip_pitch_to_melodic_range(note.pitch)
                if pitch_idx == -1:
                    continue
                grid[b, s, pitch_idx] = 1.0

    return {
        "strings": strings,
        "woodwinds": woodwinds,
        "brass": brass,
        "perc": perc,
    }


def grids_to_feature_sequences(
    family_grids: Dict[str, np.ndarray],
) -> np.ndarray:
    """
    Convert per-family grids into a [T, 39] feature sequence.

    Families:
        strings:   [bars, STEPS_PER_BAR, 12]
        woodwinds: [bars, STEPS_PER_BAR, 12]
        brass:     [bars, STEPS_PER_BAR, 12]
        perc:      [bars, STEPS_PER_BAR, 3]

    Returns:
        seq: [T, 39], where T = bars * STEPS_PER_BAR.
    """
    strings = family_grids["strings"]
    woodwinds = family_grids["woodwinds"]
    brass = family_grids["brass"]
    perc = family_grids["perc"]

    bars, steps_per_bar, n_strings = strings.shape
    T = bars * steps_per_bar

    # Reshape to [T, pitches]
    strings_flat = strings.reshape(T, n_strings)
    woodwinds_flat = woodwinds.reshape(T, woodwinds.shape[-1])
    brass_flat = brass.reshape(T, brass.shape[-1])
    perc_flat = perc.reshape(T, perc.shape[-1])

    # Concatenate along feature dimension: [T, 12+12+12+3] = [T, 39]
    seq = np.concatenate(
        [strings_flat, woodwinds_flat, brass_flat, perc_flat],
        axis=-1,
    )

    return seq


def build_sequences_for_file(
    midi_path: str,
    seq_len_steps: int,
    max_bars: int = None,
) -> Tuple[np.ndarray, np.ndarray]:
    """
    Build (X, Y) sequences for a single MIDI file.

    - Convert MIDI → family grids → [T, 39] sequence.
    - Build sliding window sequences of length seq_len_steps:
        X: first seq_len_steps-1 steps
        Y: next step (for next-step prediction)
      For convenience, we pack them as:
        X: [N, seq_len_steps, 39]
        Y: [N, seq_len_steps, 39] (shifted by 1).

    Args:
        midi_path: Path to MIDI file.
        seq_len_steps: Sequence length in grid steps (e.g., 64, 128).
        max_bars: Optional max bars to consider from this file.

    Returns:
        X: [N, seq_len_steps, 39]
        Y: [N, seq_len_steps, 39]
    """
    try:
        pm = pretty_midi.PrettyMIDI(midi_path)
    except Exception:
        # Skip unreadable files
        return np.empty((0, seq_len_steps, config.INPUT_FEATURES), dtype=np.float32), \
               np.empty((0, seq_len_steps, config.INPUT_FEATURES), dtype=np.float32)

    family_grids = instrument_to_family_grids(pm, max_bars=max_bars)
    seq = grids_to_feature_sequences(family_grids)  # [T, 39]

    T = seq.shape[0]
    if T <= seq_len_steps:
        return np.empty((0, seq_len_steps, config.INPUT_FEATURES), dtype=np.float32), \
               np.empty((0, seq_len_steps, config.INPUT_FEATURES), dtype=np.float32)

    # Sliding window
    X_list = []
    Y_list = []
    # We'll create sequences of length seq_len_steps,
    # with Y being the same sequence shifted by 1 step.
    for start in range(0, T - seq_len_steps):
        end = start + seq_len_steps
        seq_window = seq[start:end]           # [seq_len_steps, 39]
        seq_window_next = seq[start + 1:end + 1]  # shifted by 1 step

        X_list.append(seq_window)
        Y_list.append(seq_window_next)

    X = np.stack(X_list, axis=0)  # [N, seq_len_steps, 39]
    Y = np.stack(Y_list, axis=0)  # [N, seq_len_steps, 39]

    return X.astype(np.float32), Y.astype(np.float32)


def build_dataset_from_dir(
    midi_root: str,
    seq_len_bars: int,
    max_files: int = None,
    max_bars_per_file: int = None,
    pattern: str = "**/*.mid*",
) -> Tuple[np.ndarray, np.ndarray]:
    """
    Build a full dataset (X, Y) from all MIDI files in a directory.

    Args:
        midi_root: Root directory containing MIDI files (e.g., MusicNet midis).
        seq_len_bars: Sequence length in bars.
        max_files: Optional max number of files to use.
        max_bars_per_file: Optional cap on bars per file for preprocessing.
        pattern: Glob pattern for finding midis under midi_root.

    Returns:
        X_all: [N_total, seq_len_steps, 39]
        Y_all: [N_total, seq_len_steps, 39]
    """
    midi_paths = load_midi_paths_from_dir(midi_root, pattern=pattern)

    if max_files is not None:
        midi_paths = midi_paths[:max_files]

    seq_len_steps = seq_len_bars * config.STEPS_PER_BAR

    X_all_list = []
    Y_all_list = []

    for path in midi_paths:
        X, Y = build_sequences_for_file(
            midi_path=path,
            seq_len_steps=seq_len_steps,
            max_bars=max_bars_per_file,
        )
        if X.shape[0] == 0:
            continue
        X_all_list.append(X)
        Y_all_list.append(Y)

    if not X_all_list:
        # No data
        return np.empty((0, seq_len_steps, config.INPUT_FEATURES), dtype=np.float32), \
               np.empty((0, seq_len_steps, config.INPUT_FEATURES), dtype=np.float32)

    X_all = np.concatenate(X_all_list, axis=0)
    Y_all = np.concatenate(Y_all_list, axis=0)

    return X_all, Y_all


def save_dataset_npz(
    X: np.ndarray,
    Y: np.ndarray,
    out_path: str,
) -> None:
    """
    Save (X, Y) arrays to a .npz file.

    Args:
        X: [N, seq_len_steps, 39]
        Y: [N, seq_len_steps, 39]
        out_path: path to .npz file.
    """
    np.savez_compressed(out_path, X=X, Y=Y)


def save_dataset_pt(
    X: np.ndarray,
    Y: np.ndarray,
    out_path: str,
) -> None:
    """
    Save (X, Y) arrays to a .pt file for PyTorch.

    Args:
        X: [N, seq_len_steps, 39]
        Y: [N, seq_len_steps, 39]
        out_path: path to .pt file.
    """
    import torch

    data = {"X": torch.from_numpy(X), "Y": torch.from_numpy(Y)}
    torch.save(data, out_path)