Upload hv_note.py
Browse files- hv_note.py +824 -0
hv_note.py
ADDED
|
@@ -0,0 +1,824 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
"""
|
| 2 |
+
hv-note
|
| 3 |
+
=======
|
| 4 |
+
|
| 5 |
+
Symbolic audio event extraction. No lexicon. No translation layer.
|
| 6 |
+
|
| 7 |
+
Given a WAV file, output a score: a sequence of (onset_frame, symbol_id)
|
| 8 |
+
pairs, where each symbol describes a discrete audio event by
|
| 9 |
+
(pitch_band, duration_class, amplitude_class).
|
| 10 |
+
|
| 11 |
+
The symbol alphabet is fixed and unnamed. Two recordings containing the
|
| 12 |
+
same pattern produce the same score. You do not need to know what the
|
| 13 |
+
pattern means to match it.
|
| 14 |
+
|
| 15 |
+
Pure stdlib. No dependencies.
|
| 16 |
+
|
| 17 |
+
Author: zeechimp
|
| 18 |
+
License: Apache-2.0
|
| 19 |
+
"""
|
| 20 |
+
|
| 21 |
+
from __future__ import annotations
|
| 22 |
+
|
| 23 |
+
import argparse
|
| 24 |
+
import json
|
| 25 |
+
import math
|
| 26 |
+
import os
|
| 27 |
+
import sys
|
| 28 |
+
import wave
|
| 29 |
+
from array import array
|
| 30 |
+
from collections import Counter
|
| 31 |
+
from dataclasses import asdict, dataclass, fields
|
| 32 |
+
from pathlib import Path
|
| 33 |
+
from typing import Dict, List, Optional, Tuple
|
| 34 |
+
|
| 35 |
+
|
| 36 |
+
# ============================================================================
|
| 37 |
+
# Pitch band grid (log-spaced, Hz)
|
| 38 |
+
# ============================================================================
|
| 39 |
+
def _log_bands(lo: float, hi: float, n: int) -> List[float]:
|
| 40 |
+
ratio = (hi / lo) ** (1.0 / (n - 1))
|
| 41 |
+
return [lo * (ratio ** i) for i in range(n)]
|
| 42 |
+
|
| 43 |
+
|
| 44 |
+
PITCH_BANDS: List[float] = _log_bands(80.0, 6000.0, 16)
|
| 45 |
+
N_BANDS: int = len(PITCH_BANDS)
|
| 46 |
+
N_DUR: int = 4
|
| 47 |
+
N_AMP: int = 3
|
| 48 |
+
N_SYMBOLS: int = N_BANDS * N_DUR * N_AMP # 192
|
| 49 |
+
|
| 50 |
+
|
| 51 |
+
def encode_symbol(band: int, dur: int, amp: int) -> int:
|
| 52 |
+
return band * (N_DUR * N_AMP) + dur * N_AMP + amp
|
| 53 |
+
|
| 54 |
+
|
| 55 |
+
def decode_symbol(sid: int) -> Tuple[int, int, int]:
|
| 56 |
+
band = sid // (N_DUR * N_AMP)
|
| 57 |
+
rem = sid % (N_DUR * N_AMP)
|
| 58 |
+
return band, rem // N_AMP, rem % N_AMP
|
| 59 |
+
|
| 60 |
+
|
| 61 |
+
# ============================================================================
|
| 62 |
+
# Config
|
| 63 |
+
# ============================================================================
|
| 64 |
+
@dataclass
|
| 65 |
+
class HVNoteConfig:
|
| 66 |
+
# Framing
|
| 67 |
+
frame_ms: float = 20.0
|
| 68 |
+
hop_ms: float = 10.0
|
| 69 |
+
|
| 70 |
+
# Onset detection
|
| 71 |
+
onset_floor: float = 0.02
|
| 72 |
+
onset_rise_mult: float = 1.5
|
| 73 |
+
|
| 74 |
+
# Event segmentation
|
| 75 |
+
min_peak_rms: float = 0.02
|
| 76 |
+
max_event_frames: int = 200
|
| 77 |
+
min_event_frames: int = 1
|
| 78 |
+
|
| 79 |
+
# Pitch window
|
| 80 |
+
pitch_window_s: float = 0.030
|
| 81 |
+
|
| 82 |
+
# Quantization
|
| 83 |
+
dur_bounds: Tuple[int, int, int] = (3, 8, 20)
|
| 84 |
+
amp_bounds: Tuple[float, float] = (0.3, 0.7)
|
| 85 |
+
|
| 86 |
+
# Corpus
|
| 87 |
+
max_files: int = 0
|
| 88 |
+
glob: str = "*.wav"
|
| 89 |
+
|
| 90 |
+
version: str = "0.1.4"
|
| 91 |
+
|
| 92 |
+
|
| 93 |
+
# ============================================================================
|
| 94 |
+
# Report dataclasses
|
| 95 |
+
# ============================================================================
|
| 96 |
+
@dataclass
|
| 97 |
+
class AudioEvent:
|
| 98 |
+
start_frame: int
|
| 99 |
+
end_frame: int
|
| 100 |
+
peak_frame: int
|
| 101 |
+
start_s: float
|
| 102 |
+
end_s: float
|
| 103 |
+
peak_s: float
|
| 104 |
+
peak_rms: float
|
| 105 |
+
pitch_hz: float
|
| 106 |
+
band: int
|
| 107 |
+
dur_class: int
|
| 108 |
+
amp_class: int
|
| 109 |
+
symbol: int
|
| 110 |
+
|
| 111 |
+
def to_dict(self) -> dict:
|
| 112 |
+
return {
|
| 113 |
+
"start_frame": self.start_frame,
|
| 114 |
+
"end_frame": self.end_frame,
|
| 115 |
+
"peak_frame": self.peak_frame,
|
| 116 |
+
"start_s": round(self.start_s, 4),
|
| 117 |
+
"end_s": round(self.end_s, 4),
|
| 118 |
+
"peak_s": round(self.peak_s, 4),
|
| 119 |
+
"peak_rms": round(self.peak_rms, 6),
|
| 120 |
+
"pitch_hz": round(self.pitch_hz, 2),
|
| 121 |
+
"band": self.band,
|
| 122 |
+
"dur_class": self.dur_class,
|
| 123 |
+
"amp_class": self.amp_class,
|
| 124 |
+
"symbol": self.symbol,
|
| 125 |
+
}
|
| 126 |
+
|
| 127 |
+
|
| 128 |
+
@dataclass
|
| 129 |
+
class ScoreReport:
|
| 130 |
+
path: str
|
| 131 |
+
sample_rate: int
|
| 132 |
+
duration_s: float
|
| 133 |
+
n_frames: int
|
| 134 |
+
n_samples: int
|
| 135 |
+
events: List[AudioEvent]
|
| 136 |
+
score: List[Tuple[int, int]]
|
| 137 |
+
histogram: Dict[int, int]
|
| 138 |
+
entropy_bits: float
|
| 139 |
+
|
| 140 |
+
def to_dict(self) -> dict:
|
| 141 |
+
return {
|
| 142 |
+
"path": self.path,
|
| 143 |
+
"sample_rate": self.sample_rate,
|
| 144 |
+
"duration_s": round(self.duration_s, 4),
|
| 145 |
+
"n_frames": self.n_frames,
|
| 146 |
+
"n_samples": self.n_samples,
|
| 147 |
+
"n_events": len(self.events),
|
| 148 |
+
"score": [[int(t), int(s)] for t, s in self.score],
|
| 149 |
+
"histogram": {str(k): v for k, v in self.histogram.items()},
|
| 150 |
+
"entropy_bits": round(self.entropy_bits, 4),
|
| 151 |
+
"events": [e.to_dict() for e in self.events],
|
| 152 |
+
}
|
| 153 |
+
|
| 154 |
+
|
| 155 |
+
@dataclass
|
| 156 |
+
class CorpusReport:
|
| 157 |
+
root: str
|
| 158 |
+
n_files: int
|
| 159 |
+
n_events: int
|
| 160 |
+
unique_symbols: int
|
| 161 |
+
total_entropy_bits: float
|
| 162 |
+
symbol_counts: Dict[int, int]
|
| 163 |
+
group_counts: Dict[str, int]
|
| 164 |
+
group_unique: Dict[str, int]
|
| 165 |
+
file_scores: Dict[str, List[int]]
|
| 166 |
+
|
| 167 |
+
def to_dict(self) -> dict:
|
| 168 |
+
return {
|
| 169 |
+
"root": self.root,
|
| 170 |
+
"n_files": self.n_files,
|
| 171 |
+
"n_events": self.n_events,
|
| 172 |
+
"unique_symbols": self.unique_symbols,
|
| 173 |
+
"total_entropy_bits": round(self.total_entropy_bits, 4),
|
| 174 |
+
"symbol_counts": {str(k): v for k, v in self.symbol_counts.items()},
|
| 175 |
+
"group_counts": dict(self.group_counts),
|
| 176 |
+
"group_unique": dict(self.group_unique),
|
| 177 |
+
"file_scores": {k: list(v) for k, v in self.file_scores.items()},
|
| 178 |
+
}
|
| 179 |
+
|
| 180 |
+
|
| 181 |
+
# ============================================================================
|
| 182 |
+
# Audio I/O & primitives
|
| 183 |
+
# ============================================================================
|
| 184 |
+
def read_wav(path: str) -> Tuple[array, int]:
|
| 185 |
+
"""Return (samples_int16, sample_rate). Left channel only."""
|
| 186 |
+
with wave.open(path, "rb") as w:
|
| 187 |
+
n_ch = w.getnchannels()
|
| 188 |
+
sw = w.getsampwidth()
|
| 189 |
+
sr = w.getframerate()
|
| 190 |
+
n = w.getnframes()
|
| 191 |
+
raw = w.readframes(n)
|
| 192 |
+
|
| 193 |
+
if sw != 2:
|
| 194 |
+
raise ValueError(f"Only 16-bit PCM WAV supported, got {sw*8}-bit")
|
| 195 |
+
|
| 196 |
+
samples = array("h")
|
| 197 |
+
samples.frombytes(raw)
|
| 198 |
+
|
| 199 |
+
if n_ch > 1:
|
| 200 |
+
samples = samples[::n_ch]
|
| 201 |
+
|
| 202 |
+
return samples, sr
|
| 203 |
+
|
| 204 |
+
|
| 205 |
+
def rms_norm(chunk: array) -> float:
|
| 206 |
+
n = len(chunk)
|
| 207 |
+
if n == 0:
|
| 208 |
+
return 0.0
|
| 209 |
+
s = 0
|
| 210 |
+
for x in chunk:
|
| 211 |
+
s += x * x
|
| 212 |
+
return math.sqrt(s / n) / 32768.0
|
| 213 |
+
|
| 214 |
+
|
| 215 |
+
def goertzel_power(window: array, target_hz: float, sr: int) -> float:
|
| 216 |
+
n = len(window)
|
| 217 |
+
if n == 0 or target_hz >= sr / 2:
|
| 218 |
+
return 0.0
|
| 219 |
+
k = int(0.5 + n * target_hz / sr)
|
| 220 |
+
if k < 1:
|
| 221 |
+
k = 1
|
| 222 |
+
omega = 2.0 * math.pi * k / n
|
| 223 |
+
coeff = 2.0 * math.cos(omega)
|
| 224 |
+
s_prev = 0.0
|
| 225 |
+
s_prev2 = 0.0
|
| 226 |
+
for x in window:
|
| 227 |
+
s = x + coeff * s_prev - s_prev2
|
| 228 |
+
s_prev2 = s_prev
|
| 229 |
+
s_prev = s
|
| 230 |
+
return s_prev2 * s_prev2 + s_prev * s_prev - coeff * s_prev * s_prev2
|
| 231 |
+
|
| 232 |
+
|
| 233 |
+
def entropy_bits(counts: Dict[int, int]) -> float:
|
| 234 |
+
total = sum(counts.values())
|
| 235 |
+
if total == 0:
|
| 236 |
+
return 0.0
|
| 237 |
+
e = 0.0
|
| 238 |
+
for c in counts.values():
|
| 239 |
+
if c > 0:
|
| 240 |
+
p = c / total
|
| 241 |
+
e -= p * math.log2(p)
|
| 242 |
+
return e
|
| 243 |
+
|
| 244 |
+
|
| 245 |
+
# ============================================================================
|
| 246 |
+
# The model
|
| 247 |
+
# ============================================================================
|
| 248 |
+
class HVNote:
|
| 249 |
+
"""Symbolic audio event extraction."""
|
| 250 |
+
|
| 251 |
+
def __init__(self, config: Optional[HVNoteConfig] = None):
|
| 252 |
+
self.config = config or HVNoteConfig()
|
| 253 |
+
self._obs = 0
|
| 254 |
+
|
| 255 |
+
def __repr__(self) -> str:
|
| 256 |
+
return (
|
| 257 |
+
f"HVNote(frame_ms={self.config.frame_ms}, "
|
| 258 |
+
f"hop_ms={self.config.hop_ms}, "
|
| 259 |
+
f"n_symbols={N_SYMBOLS}, "
|
| 260 |
+
f"version={self.config.version})"
|
| 261 |
+
)
|
| 262 |
+
|
| 263 |
+
# ------------------------------------------------------------------
|
| 264 |
+
# Framing
|
| 265 |
+
# ------------------------------------------------------------------
|
| 266 |
+
def _frame_rms(self, samples: array, sr: int) -> Tuple[List[float], int, int]:
|
| 267 |
+
win = max(1, int(sr * self.config.frame_ms / 1000))
|
| 268 |
+
hop = max(1, int(sr * self.config.hop_ms / 1000))
|
| 269 |
+
rms_vals: List[float] = []
|
| 270 |
+
i = 0
|
| 271 |
+
n = len(samples)
|
| 272 |
+
while i + win <= n:
|
| 273 |
+
rms_vals.append(rms_norm(samples[i:i + win]))
|
| 274 |
+
i += hop
|
| 275 |
+
return rms_vals, win, hop
|
| 276 |
+
|
| 277 |
+
# ------------------------------------------------------------------
|
| 278 |
+
# Event detection
|
| 279 |
+
# ------------------------------------------------------------------
|
| 280 |
+
def _detect_events(
|
| 281 |
+
self, rms_vals: List[float]
|
| 282 |
+
) -> List[Tuple[int, int, int, float]]:
|
| 283 |
+
"""
|
| 284 |
+
Onset + extent in a single forward scan.
|
| 285 |
+
|
| 286 |
+
Onset fires at frame i when rms[i] >= onset_floor AND either:
|
| 287 |
+
- rms[i] >= rms[i-1] * onset_rise_mult (sharp rise), or
|
| 288 |
+
- rms[i-1] < onset_floor (silence crossing).
|
| 289 |
+
|
| 290 |
+
The event's extent is the connected region above onset_floor
|
| 291 |
+
starting at the onset. The peak is argmax over that region only —
|
| 292 |
+
never over a fixed search horizon. This prevents a later, louder
|
| 293 |
+
event from hijacking the current one.
|
| 294 |
+
"""
|
| 295 |
+
n = len(rms_vals)
|
| 296 |
+
if n == 0:
|
| 297 |
+
return []
|
| 298 |
+
|
| 299 |
+
floor = self.config.onset_floor
|
| 300 |
+
rise_mult = self.config.onset_rise_mult
|
| 301 |
+
|
| 302 |
+
events: List[Tuple[int, int, int, float]] = []
|
| 303 |
+
i = 0
|
| 304 |
+
while i < n:
|
| 305 |
+
cur = rms_vals[i]
|
| 306 |
+
if cur < floor:
|
| 307 |
+
i += 1
|
| 308 |
+
continue
|
| 309 |
+
|
| 310 |
+
prev = rms_vals[i - 1] if i > 0 else 0.0
|
| 311 |
+
is_rise = (i == 0) or (prev > 1e-9 and cur >= prev * rise_mult)
|
| 312 |
+
is_crossing = prev < floor
|
| 313 |
+
if not (is_rise or is_crossing):
|
| 314 |
+
i += 1
|
| 315 |
+
continue
|
| 316 |
+
|
| 317 |
+
# Extent: stay in the connected region above floor.
|
| 318 |
+
max_end = min(n, i + self.config.max_event_frames)
|
| 319 |
+
end = i
|
| 320 |
+
while end < max_end and rms_vals[end] >= floor:
|
| 321 |
+
end += 1
|
| 322 |
+
|
| 323 |
+
if end - i < self.config.min_event_frames:
|
| 324 |
+
i += 1
|
| 325 |
+
continue
|
| 326 |
+
|
| 327 |
+
# Peak is argmax over the extent only.
|
| 328 |
+
peak_frame = max(range(i, end), key=lambda j: rms_vals[j])
|
| 329 |
+
peak_rms = rms_vals[peak_frame]
|
| 330 |
+
if peak_rms < self.config.min_peak_rms:
|
| 331 |
+
i += 1
|
| 332 |
+
continue
|
| 333 |
+
|
| 334 |
+
events.append((i, end, peak_frame, peak_rms))
|
| 335 |
+
# Skip past the whole event.
|
| 336 |
+
i = max(i + 1, end)
|
| 337 |
+
|
| 338 |
+
return events
|
| 339 |
+
|
| 340 |
+
# ------------------------------------------------------------------
|
| 341 |
+
# Pitch band
|
| 342 |
+
# ------------------------------------------------------------------
|
| 343 |
+
def _pitch_band(
|
| 344 |
+
self, samples: array, sr: int, peak_frame: int, hop: int
|
| 345 |
+
) -> Tuple[int, float]:
|
| 346 |
+
window_len = max(1, int(sr * self.config.pitch_window_s))
|
| 347 |
+
center = peak_frame * hop + (hop // 2)
|
| 348 |
+
start = max(0, center - window_len // 2)
|
| 349 |
+
end = min(len(samples), start + window_len)
|
| 350 |
+
window = samples[start:end]
|
| 351 |
+
if len(window) < 8:
|
| 352 |
+
return 0, 0.0
|
| 353 |
+
|
| 354 |
+
best_band = 0
|
| 355 |
+
best_power = -1.0
|
| 356 |
+
for bi, f in enumerate(PITCH_BANDS):
|
| 357 |
+
p = goertzel_power(window, f, sr)
|
| 358 |
+
if p > best_power:
|
| 359 |
+
best_power = p
|
| 360 |
+
best_band = bi
|
| 361 |
+
return best_band, PITCH_BANDS[best_band]
|
| 362 |
+
|
| 363 |
+
# ------------------------------------------------------------------
|
| 364 |
+
# Quantization
|
| 365 |
+
# ------------------------------------------------------------------
|
| 366 |
+
def _dur_class(self, length: int) -> int:
|
| 367 |
+
b0, b1, b2 = self.config.dur_bounds
|
| 368 |
+
if length < b0:
|
| 369 |
+
return 0
|
| 370 |
+
if length < b1:
|
| 371 |
+
return 1
|
| 372 |
+
if length < b2:
|
| 373 |
+
return 2
|
| 374 |
+
return 3
|
| 375 |
+
|
| 376 |
+
def _amp_class(self, peak_rms: float, recording_peak: float) -> int:
|
| 377 |
+
if recording_peak <= 0:
|
| 378 |
+
return 0
|
| 379 |
+
r = peak_rms / recording_peak
|
| 380 |
+
a0, a1 = self.config.amp_bounds
|
| 381 |
+
if r < a0:
|
| 382 |
+
return 0
|
| 383 |
+
if r < a1:
|
| 384 |
+
return 1
|
| 385 |
+
return 2
|
| 386 |
+
|
| 387 |
+
# ------------------------------------------------------------------
|
| 388 |
+
# Public API
|
| 389 |
+
# ------------------------------------------------------------------
|
| 390 |
+
def score_file(self, path: str) -> ScoreReport:
|
| 391 |
+
samples, sr = read_wav(path)
|
| 392 |
+
return self.score_samples(samples, sr, path=path)
|
| 393 |
+
|
| 394 |
+
def score_samples(
|
| 395 |
+
self, samples: array, sr: int, path: str = "<memory>"
|
| 396 |
+
) -> ScoreReport:
|
| 397 |
+
rms_vals, win, hop = self._frame_rms(samples, sr)
|
| 398 |
+
n_frames = len(rms_vals)
|
| 399 |
+
recording_peak = max(rms_vals) if rms_vals else 0.0
|
| 400 |
+
|
| 401 |
+
raw_events = self._detect_events(rms_vals)
|
| 402 |
+
|
| 403 |
+
events: List[AudioEvent] = []
|
| 404 |
+
for start, end, peak_frame, peak_rms in raw_events:
|
| 405 |
+
band, hz = self._pitch_band(samples, sr, peak_frame, hop)
|
| 406 |
+
dur = self._dur_class(end - start)
|
| 407 |
+
amp = self._amp_class(peak_rms, recording_peak)
|
| 408 |
+
sym = encode_symbol(band, dur, amp)
|
| 409 |
+
events.append(AudioEvent(
|
| 410 |
+
start_frame=start,
|
| 411 |
+
end_frame=end,
|
| 412 |
+
peak_frame=peak_frame,
|
| 413 |
+
start_s=start * self.config.hop_ms / 1000.0,
|
| 414 |
+
end_s=end * self.config.hop_ms / 1000.0,
|
| 415 |
+
peak_s=peak_frame * self.config.hop_ms / 1000.0,
|
| 416 |
+
peak_rms=peak_rms,
|
| 417 |
+
pitch_hz=hz,
|
| 418 |
+
band=band,
|
| 419 |
+
dur_class=dur,
|
| 420 |
+
amp_class=amp,
|
| 421 |
+
symbol=sym,
|
| 422 |
+
))
|
| 423 |
+
|
| 424 |
+
score = [(e.start_frame, e.symbol) for e in events]
|
| 425 |
+
hist = Counter(e.symbol for e in events)
|
| 426 |
+
hist_dict = {int(k): int(v) for k, v in hist.items()}
|
| 427 |
+
ent = entropy_bits(hist_dict)
|
| 428 |
+
dur = len(samples) / sr if sr > 0 else 0.0
|
| 429 |
+
|
| 430 |
+
self._obs += 1
|
| 431 |
+
return ScoreReport(
|
| 432 |
+
path=path,
|
| 433 |
+
sample_rate=sr,
|
| 434 |
+
duration_s=dur,
|
| 435 |
+
n_frames=n_frames,
|
| 436 |
+
n_samples=len(samples),
|
| 437 |
+
events=events,
|
| 438 |
+
score=score,
|
| 439 |
+
histogram=hist_dict,
|
| 440 |
+
entropy_bits=ent,
|
| 441 |
+
)
|
| 442 |
+
|
| 443 |
+
def score_directory(
|
| 444 |
+
self, root: str, max_files: Optional[int] = None
|
| 445 |
+
) -> CorpusReport:
|
| 446 |
+
root_path = Path(root)
|
| 447 |
+
if not root_path.exists():
|
| 448 |
+
raise FileNotFoundError(root)
|
| 449 |
+
|
| 450 |
+
files = sorted(root_path.rglob(self.config.glob))
|
| 451 |
+
if not files:
|
| 452 |
+
raise FileNotFoundError(
|
| 453 |
+
f"No files matching {self.config.glob!r} under {root}"
|
| 454 |
+
)
|
| 455 |
+
|
| 456 |
+
limit = max_files if max_files is not None else self.config.max_files
|
| 457 |
+
if limit and limit > 0:
|
| 458 |
+
files = files[:limit]
|
| 459 |
+
|
| 460 |
+
symbol_counts: Counter = Counter()
|
| 461 |
+
group_counts: Counter = Counter()
|
| 462 |
+
group_symbols: Dict[str, set] = {}
|
| 463 |
+
file_scores: Dict[str, List[int]] = {}
|
| 464 |
+
total_events = 0
|
| 465 |
+
|
| 466 |
+
for f in files:
|
| 467 |
+
try:
|
| 468 |
+
report = self.score_file(str(f))
|
| 469 |
+
except Exception as e:
|
| 470 |
+
print(f" skip {f.name}: {e}", file=sys.stderr)
|
| 471 |
+
continue
|
| 472 |
+
|
| 473 |
+
file_scores[str(f)] = [s for _, s in report.score]
|
| 474 |
+
total_events += len(report.score)
|
| 475 |
+
|
| 476 |
+
group_key = self._group_key(f.stem)
|
| 477 |
+
group_counts[group_key] += 1
|
| 478 |
+
group_symbols.setdefault(group_key, set())
|
| 479 |
+
group_symbols[group_key].update(s for _, s in report.score)
|
| 480 |
+
symbol_counts.update(s for _, s in report.score)
|
| 481 |
+
|
| 482 |
+
symbol_counts_dict = {int(k): int(v) for k, v in symbol_counts.items()}
|
| 483 |
+
group_unique = {k: len(v) for k, v in group_symbols.items()}
|
| 484 |
+
ent = entropy_bits(symbol_counts_dict)
|
| 485 |
+
|
| 486 |
+
return CorpusReport(
|
| 487 |
+
root=str(root),
|
| 488 |
+
n_files=len(file_scores),
|
| 489 |
+
n_events=total_events,
|
| 490 |
+
unique_symbols=len(symbol_counts),
|
| 491 |
+
total_entropy_bits=ent,
|
| 492 |
+
symbol_counts=symbol_counts_dict,
|
| 493 |
+
group_counts={k: int(v) for k, v in group_counts.items()},
|
| 494 |
+
group_unique=group_unique,
|
| 495 |
+
file_scores=file_scores,
|
| 496 |
+
)
|
| 497 |
+
|
| 498 |
+
@staticmethod
|
| 499 |
+
def _group_key(stem: str) -> str:
|
| 500 |
+
for sep in ("_", "-"):
|
| 501 |
+
if sep in stem:
|
| 502 |
+
return stem.split(sep)[0]
|
| 503 |
+
return stem
|
| 504 |
+
|
| 505 |
+
# ------------------------------------------------------------------
|
| 506 |
+
# Convenience
|
| 507 |
+
# ------------------------------------------------------------------
|
| 508 |
+
def score(self, path: str) -> List[int]:
|
| 509 |
+
return [s for _, s in self.score_file(path).score]
|
| 510 |
+
|
| 511 |
+
def fingerprint(self, path: str, k: int = 8) -> Tuple[int, ...]:
|
| 512 |
+
s = self.score(path)
|
| 513 |
+
return tuple(s[:k])
|
| 514 |
+
|
| 515 |
+
# ------------------------------------------------------------------
|
| 516 |
+
# Rendering
|
| 517 |
+
# ------------------------------------------------------------------
|
| 518 |
+
def render_file(self, report: ScoreReport) -> str:
|
| 519 |
+
lines: List[str] = []
|
| 520 |
+
bar = "=" * 72
|
| 521 |
+
lines.append(bar)
|
| 522 |
+
lines.append("hv-note — symbolic audio score")
|
| 523 |
+
lines.append(bar)
|
| 524 |
+
lines.append("")
|
| 525 |
+
lines.append(f" file : {report.path}")
|
| 526 |
+
lines.append(f" duration : {report.duration_s:.3f} s")
|
| 527 |
+
lines.append(f" sample rate: {report.sample_rate} Hz")
|
| 528 |
+
lines.append(f" frames : {report.n_frames}")
|
| 529 |
+
lines.append(f" events : {len(report.events)}")
|
| 530 |
+
lines.append(f" unique sym : {len(report.histogram)}")
|
| 531 |
+
lines.append(f" entropy : {report.entropy_bits:.3f} bits")
|
| 532 |
+
lines.append("")
|
| 533 |
+
|
| 534 |
+
if not report.events:
|
| 535 |
+
lines.append(" (no events detected)")
|
| 536 |
+
return "\n".join(lines)
|
| 537 |
+
|
| 538 |
+
lines.append(" SCORE")
|
| 539 |
+
lines.append(" " + "-" * 68)
|
| 540 |
+
lines.append(
|
| 541 |
+
f" {'frame':>6} {'time':>7} {'sym':>4} "
|
| 542 |
+
f"(band, dur, amp) {'pitch':>8} rms"
|
| 543 |
+
)
|
| 544 |
+
for e in report.events:
|
| 545 |
+
lines.append(
|
| 546 |
+
f" {e.start_frame:>6} {e.start_s:>6.3f}s "
|
| 547 |
+
f"{e.symbol:>4} "
|
| 548 |
+
f"({e.band:>2}, {e.dur_class}, {e.amp_class}) "
|
| 549 |
+
f"{e.pitch_hz:>7.1f}Hz {e.peak_rms:.4f}"
|
| 550 |
+
)
|
| 551 |
+
lines.append("")
|
| 552 |
+
|
| 553 |
+
lines.append(" HISTOGRAM")
|
| 554 |
+
lines.append(" " + "-" * 68)
|
| 555 |
+
for sym, count in sorted(
|
| 556 |
+
report.histogram.items(), key=lambda kv: -kv[1]
|
| 557 |
+
):
|
| 558 |
+
band, dur, amp = decode_symbol(sym)
|
| 559 |
+
lines.append(
|
| 560 |
+
f" {sym:>4} ({band:>2}, {dur}, {amp}) {count:>4}"
|
| 561 |
+
)
|
| 562 |
+
lines.append("")
|
| 563 |
+
|
| 564 |
+
if len(report.score) >= 2:
|
| 565 |
+
lines.append(" TRANSITIONS (top 8)")
|
| 566 |
+
lines.append(" " + "-" * 68)
|
| 567 |
+
trans: Counter = Counter()
|
| 568 |
+
for i in range(len(report.score) - 1):
|
| 569 |
+
a = report.score[i][1]
|
| 570 |
+
b = report.score[i + 1][1]
|
| 571 |
+
trans[(a, b)] += 1
|
| 572 |
+
for (a, b), c in trans.most_common(8):
|
| 573 |
+
lines.append(f" {a:>4} -> {b:<4} {c:>4}")
|
| 574 |
+
lines.append("")
|
| 575 |
+
|
| 576 |
+
return "\n".join(lines)
|
| 577 |
+
|
| 578 |
+
def render_corpus(self, report: CorpusReport) -> str:
|
| 579 |
+
lines: List[str] = []
|
| 580 |
+
bar = "=" * 72
|
| 581 |
+
lines.append(bar)
|
| 582 |
+
lines.append("hv-note — corpus of symbolic audio scores")
|
| 583 |
+
lines.append(bar)
|
| 584 |
+
lines.append("")
|
| 585 |
+
lines.append(f" root : {report.root}")
|
| 586 |
+
lines.append(f" files : {report.n_files}")
|
| 587 |
+
lines.append(f" events : {report.n_events}")
|
| 588 |
+
lines.append(f" unique sym : {report.unique_symbols} / {N_SYMBOLS}")
|
| 589 |
+
lines.append(f" entropy : {report.total_entropy_bits:.3f} bits")
|
| 590 |
+
lines.append("")
|
| 591 |
+
|
| 592 |
+
if report.symbol_counts:
|
| 593 |
+
lines.append(" TOP SYMBOLS")
|
| 594 |
+
lines.append(" " + "-" * 68)
|
| 595 |
+
lines.append(
|
| 596 |
+
f" {'sym':>4} (band, dur, amp) count"
|
| 597 |
+
)
|
| 598 |
+
for sym, count in sorted(
|
| 599 |
+
report.symbol_counts.items(), key=lambda kv: -kv[1]
|
| 600 |
+
)[:15]:
|
| 601 |
+
band, dur, amp = decode_symbol(sym)
|
| 602 |
+
lines.append(
|
| 603 |
+
f" {sym:>4} ({band:>2}, {dur}, {amp}) "
|
| 604 |
+
f"{count:>6}"
|
| 605 |
+
)
|
| 606 |
+
lines.append("")
|
| 607 |
+
|
| 608 |
+
if report.group_counts:
|
| 609 |
+
lines.append(" PER-GROUP SUMMARY")
|
| 610 |
+
lines.append(" " + "-" * 68)
|
| 611 |
+
lines.append(
|
| 612 |
+
f" {'group':<20} {'files':>6} {'unique_sym':>10}"
|
| 613 |
+
)
|
| 614 |
+
for g, count in sorted(
|
| 615 |
+
report.group_counts.items(), key=lambda kv: -kv[1]
|
| 616 |
+
)[:20]:
|
| 617 |
+
uniq = report.group_unique.get(g, 0)
|
| 618 |
+
lines.append(f" {g:<20} {count:>6} {uniq:>10}")
|
| 619 |
+
lines.append("")
|
| 620 |
+
|
| 621 |
+
return "\n".join(lines)
|
| 622 |
+
|
| 623 |
+
# ------------------------------------------------------------------
|
| 624 |
+
# Persistence
|
| 625 |
+
# ------------------------------------------------------------------
|
| 626 |
+
def save_pretrained(self, save_dir: str) -> None:
|
| 627 |
+
os.makedirs(save_dir, exist_ok=True)
|
| 628 |
+
payload = {
|
| 629 |
+
"config": asdict(self.config),
|
| 630 |
+
"observations": self._obs,
|
| 631 |
+
"pitch_bands": list(PITCH_BANDS),
|
| 632 |
+
"n_symbols": N_SYMBOLS,
|
| 633 |
+
}
|
| 634 |
+
with open(os.path.join(save_dir, "config.json"), "w") as f:
|
| 635 |
+
json.dump(payload, f, indent=2)
|
| 636 |
+
|
| 637 |
+
@classmethod
|
| 638 |
+
def from_pretrained(cls, save_dir: str) -> "HVNote":
|
| 639 |
+
with open(os.path.join(save_dir, "config.json"), "r") as f:
|
| 640 |
+
payload = json.load(f)
|
| 641 |
+
cfg_dict = payload.get("config", {})
|
| 642 |
+
known = {f.name for f in fields(HVNoteConfig)}
|
| 643 |
+
cfg_dict = {k: v for k, v in cfg_dict.items() if k in known}
|
| 644 |
+
cfg = HVNoteConfig(**cfg_dict)
|
| 645 |
+
obj = cls(config=cfg)
|
| 646 |
+
obj._obs = int(payload.get("observations", 0))
|
| 647 |
+
return obj
|
| 648 |
+
|
| 649 |
+
|
| 650 |
+
# ============================================================================
|
| 651 |
+
# Demo
|
| 652 |
+
# ============================================================================
|
| 653 |
+
def _write_wav(path: str, samples: List[float], sr: int) -> None:
|
| 654 |
+
ints = array(
|
| 655 |
+
"h",
|
| 656 |
+
(max(-32768, min(32767, int(x * 32767))) for x in samples),
|
| 657 |
+
)
|
| 658 |
+
with wave.open(path, "wb") as w:
|
| 659 |
+
w.setnchannels(1)
|
| 660 |
+
w.setsampwidth(2)
|
| 661 |
+
w.setframerate(sr)
|
| 662 |
+
w.writeframes(ints.tobytes())
|
| 663 |
+
|
| 664 |
+
|
| 665 |
+
def _synth_beeps(freqs: List[float], dur_ms: int, gap_ms: int,
|
| 666 |
+
sr: int = 16000) -> List[float]:
|
| 667 |
+
out: List[float] = []
|
| 668 |
+
for f in freqs:
|
| 669 |
+
n = int(sr * dur_ms / 1000)
|
| 670 |
+
for i in range(n):
|
| 671 |
+
t = i / sr
|
| 672 |
+
env = math.sin(math.pi * i / n) ** 0.5
|
| 673 |
+
out.append(0.5 * env * math.sin(2 * math.pi * f * t))
|
| 674 |
+
out.extend([0.0] * int(sr * gap_ms / 1000))
|
| 675 |
+
return out
|
| 676 |
+
|
| 677 |
+
|
| 678 |
+
def _synth_chirp(f0: float, f1: float, dur_ms: int,
|
| 679 |
+
sr: int = 16000) -> List[float]:
|
| 680 |
+
n = int(sr * dur_ms / 1000)
|
| 681 |
+
dur_s = dur_ms / 1000.0
|
| 682 |
+
out: List[float] = []
|
| 683 |
+
for i in range(n):
|
| 684 |
+
t = i / sr
|
| 685 |
+
phase = 2 * math.pi * (f0 * t + 0.5 * (f1 - f0) * t * t / dur_s)
|
| 686 |
+
env = math.sin(math.pi * i / n)
|
| 687 |
+
out.append(0.5 * env * math.sin(phase))
|
| 688 |
+
return out
|
| 689 |
+
|
| 690 |
+
|
| 691 |
+
def _synth_claps(dur_ms: int, gap_ms: int, seed: int = 42,
|
| 692 |
+
sr: int = 16000) -> List[float]:
|
| 693 |
+
import random
|
| 694 |
+
rng = random.Random(seed)
|
| 695 |
+
out: List[float] = []
|
| 696 |
+
for _ in range(4):
|
| 697 |
+
n = int(sr * dur_ms / 1000)
|
| 698 |
+
for i in range(n):
|
| 699 |
+
env = math.exp(-i / (n * 0.3))
|
| 700 |
+
out.append(0.7 * env * (rng.random() * 2 - 1))
|
| 701 |
+
out.extend([0.0] * int(sr * gap_ms / 1000))
|
| 702 |
+
return out
|
| 703 |
+
|
| 704 |
+
|
| 705 |
+
def _demo(output_dir: str = "./hv_note_output") -> None:
|
| 706 |
+
os.makedirs(output_dir, exist_ok=True)
|
| 707 |
+
m = HVNote()
|
| 708 |
+
|
| 709 |
+
samples: List[Tuple[str, List[float], int]] = [
|
| 710 |
+
("beeps_low", _synth_beeps([220, 220, 220], 150, 100), 16000),
|
| 711 |
+
("beeps_high", _synth_beeps([880, 660, 440], 120, 90), 16000),
|
| 712 |
+
("chirp_up", _synth_chirp(200, 2000, 800), 16000),
|
| 713 |
+
("claps", _synth_claps(60, 180), 16000),
|
| 714 |
+
]
|
| 715 |
+
|
| 716 |
+
print()
|
| 717 |
+
print("=" * 72)
|
| 718 |
+
print("hv-note — demo on synthesized audio")
|
| 719 |
+
print("=" * 72)
|
| 720 |
+
print()
|
| 721 |
+
|
| 722 |
+
paths: List[str] = []
|
| 723 |
+
for name, sig, sr in samples:
|
| 724 |
+
path = os.path.join(output_dir, f"{name}.wav")
|
| 725 |
+
_write_wav(path, sig, sr)
|
| 726 |
+
paths.append(path)
|
| 727 |
+
print(f" wrote {path} ({len(sig)} samples @ {sr} Hz)")
|
| 728 |
+
|
| 729 |
+
print()
|
| 730 |
+
for path in paths:
|
| 731 |
+
report = m.score_file(path)
|
| 732 |
+
print(m.render_file(report))
|
| 733 |
+
print()
|
| 734 |
+
|
| 735 |
+
print("=" * 72)
|
| 736 |
+
print("Summary")
|
| 737 |
+
print("=" * 72)
|
| 738 |
+
print(f" {'file':<24} {'events':>6} {'unique':>6} {'entropy':>8}")
|
| 739 |
+
print(" " + "-" * 60)
|
| 740 |
+
for path in paths:
|
| 741 |
+
r = m.score_file(path)
|
| 742 |
+
print(
|
| 743 |
+
f" {Path(path).name:<24} {len(r.events):>6} "
|
| 744 |
+
f"{len(r.histogram):>6} {r.entropy_bits:>7.3f}b"
|
| 745 |
+
)
|
| 746 |
+
|
| 747 |
+
print()
|
| 748 |
+
print("=" * 72)
|
| 749 |
+
print("Corpus mode (over the demo folder)")
|
| 750 |
+
print("=" * 72)
|
| 751 |
+
corpus = m.score_directory(output_dir)
|
| 752 |
+
print(m.render_corpus(corpus))
|
| 753 |
+
|
| 754 |
+
print("=" * 72)
|
| 755 |
+
print("Save / load round trip")
|
| 756 |
+
print("=" * 72)
|
| 757 |
+
save_path = os.path.join(output_dir, "hv_note_model")
|
| 758 |
+
m.save_pretrained(save_path)
|
| 759 |
+
m2 = HVNote.from_pretrained(save_path)
|
| 760 |
+
print(f" saved to : {save_path}")
|
| 761 |
+
print(f" reloaded : {m2!r}")
|
| 762 |
+
a = m.score(paths[0])
|
| 763 |
+
b = m2.score(paths[0])
|
| 764 |
+
print(f" score [0] : {a}")
|
| 765 |
+
print(f" reloaded : {b}")
|
| 766 |
+
print(f" identical : {a == b}")
|
| 767 |
+
print()
|
| 768 |
+
|
| 769 |
+
|
| 770 |
+
# ============================================================================
|
| 771 |
+
# CLI
|
| 772 |
+
# ============================================================================
|
| 773 |
+
def _cli() -> None:
|
| 774 |
+
p = argparse.ArgumentParser(
|
| 775 |
+
description="hv-note: symbolic audio event extraction."
|
| 776 |
+
)
|
| 777 |
+
p.add_argument("--file", type=str, default="",
|
| 778 |
+
help="score a single WAV file")
|
| 779 |
+
p.add_argument("--dir", type=str, default="",
|
| 780 |
+
help="score every WAV under a directory")
|
| 781 |
+
p.add_argument("--limit", type=int, default=0,
|
| 782 |
+
help="max files in --dir mode (0 = all)")
|
| 783 |
+
p.add_argument("--json", action="store_true",
|
| 784 |
+
help="output JSON instead of a rendered report")
|
| 785 |
+
p.add_argument("--fingerprint", action="store_true",
|
| 786 |
+
help="print only the fingerprint of a single file")
|
| 787 |
+
p.add_argument("--save-to", type=str, default="",
|
| 788 |
+
help="save the model to this directory")
|
| 789 |
+
p.add_argument("--outdir", type=str, default="./hv_note_output")
|
| 790 |
+
args = p.parse_args()
|
| 791 |
+
|
| 792 |
+
m = HVNote()
|
| 793 |
+
if args.save_to:
|
| 794 |
+
m.save_pretrained(args.save_to)
|
| 795 |
+
print(f"saved to {args.save_to}", file=sys.stderr)
|
| 796 |
+
|
| 797 |
+
if args.fingerprint:
|
| 798 |
+
if not args.file:
|
| 799 |
+
p.error("--fingerprint requires --file")
|
| 800 |
+
fp = m.fingerprint(args.file)
|
| 801 |
+
print(json.dumps(list(fp)))
|
| 802 |
+
return
|
| 803 |
+
|
| 804 |
+
if args.file:
|
| 805 |
+
report = m.score_file(args.file)
|
| 806 |
+
if args.json:
|
| 807 |
+
print(json.dumps(report.to_dict(), indent=2))
|
| 808 |
+
else:
|
| 809 |
+
print(m.render_file(report))
|
| 810 |
+
return
|
| 811 |
+
|
| 812 |
+
if args.dir:
|
| 813 |
+
corpus = m.score_directory(args.dir, max_files=args.limit or None)
|
| 814 |
+
if args.json:
|
| 815 |
+
print(json.dumps(corpus.to_dict(), indent=2))
|
| 816 |
+
else:
|
| 817 |
+
print(m.render_corpus(corpus))
|
| 818 |
+
return
|
| 819 |
+
|
| 820 |
+
_demo(args.outdir)
|
| 821 |
+
|
| 822 |
+
|
| 823 |
+
if __name__ == "__main__":
|
| 824 |
+
_cli()
|