Spaces:
Sleeping
Sleeping
File size: 9,032 Bytes
13e4427 | 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 | """
Phase 2 — Audio Preprocessing
Bandpass filter, wavelet denoising, normalization, silence trimming, segmentation.
"""
import logging
from pathlib import Path
from typing import Optional
import numpy as np
import pandas as pd
import librosa
import soundfile as sf
from scipy.signal import butter, sosfiltfilt
import pywt
from src.config import (
SAMPLE_RATE, BANDPASS_LOW, BANDPASS_HIGH, BANDPASS_ORDER,
WAVELET_NAME, WAVELET_LEVEL,
SEGMENT_DURATION, SEGMENT_SAMPLES, SEGMENT_HOP,
SEGMENTS_DIR, OUTPUT_DIR, LABEL_MAP,
)
logger = logging.getLogger(__name__)
def load_wav(wav_path: str, target_sr: int = SAMPLE_RATE) -> np.ndarray:
"""Load a .wav file and resample if needed.
Args:
wav_path: Path to .wav file.
target_sr: Target sample rate.
Returns:
1D numpy array of audio samples (float32).
Raises:
FileNotFoundError: If file doesn't exist.
RuntimeError: If file can't be read.
"""
path = Path(wav_path)
if not path.exists():
raise FileNotFoundError(f"WAV file not found: {wav_path}")
try:
audio, sr = librosa.load(str(wav_path), sr=target_sr, mono=True)
except Exception as e:
raise RuntimeError(f"Failed to load {wav_path}: {e}") from e
return audio.astype(np.float32)
def bandpass_filter(
audio: np.ndarray,
low: int = BANDPASS_LOW,
high: int = BANDPASS_HIGH,
sr: int = SAMPLE_RATE,
order: int = BANDPASS_ORDER,
) -> np.ndarray:
"""Apply a Butterworth bandpass filter to isolate heart sounds.
Args:
audio: 1D audio signal.
low: Low cutoff frequency (Hz).
high: High cutoff frequency (Hz).
sr: Sample rate (Hz).
order: Filter order.
Returns:
Filtered audio signal (same length).
"""
if len(audio) == 0:
return audio
nyquist = sr / 2.0
low_norm = low / nyquist
high_norm = high / nyquist
# Clamp to valid range
low_norm = max(low_norm, 1e-5)
high_norm = min(high_norm, 1.0 - 1e-5)
sos = butter(order, [low_norm, high_norm], btype="band", output="sos")
filtered = sosfiltfilt(sos, audio).astype(np.float32)
return filtered
def wavelet_denoise(
audio: np.ndarray,
wavelet: str = WAVELET_NAME,
level: int = WAVELET_LEVEL,
) -> np.ndarray:
"""Remove high-frequency noise using wavelet soft thresholding.
Args:
audio: 1D audio signal.
wavelet: Wavelet family name.
level: Decomposition level.
Returns:
Denoised audio signal.
"""
if len(audio) == 0:
return audio
# Decompose
coeffs = pywt.wavedec(audio, wavelet, level=level)
# Estimate noise from finest detail coefficients
sigma = np.median(np.abs(coeffs[-1])) / 0.6745
threshold = sigma * np.sqrt(2 * np.log(len(audio)))
# Soft-threshold detail coefficients (keep approximation intact)
denoised_coeffs = [coeffs[0]] # approximation — untouched
for detail in coeffs[1:]:
denoised_coeffs.append(pywt.threshold(detail, threshold, mode="soft"))
# Reconstruct
reconstructed = pywt.waverec(denoised_coeffs, wavelet)
# waverec can add one extra sample due to rounding
reconstructed = reconstructed[: len(audio)]
return reconstructed.astype(np.float32)
def normalize_amplitude(audio: np.ndarray) -> np.ndarray:
"""Normalize audio to [-1, 1] range by peak amplitude.
Args:
audio: 1D audio signal.
Returns:
Normalized audio. Returns zeros if input is silent.
"""
peak = np.max(np.abs(audio))
if peak < 1e-10:
return np.zeros_like(audio)
return (audio / peak).astype(np.float32)
def trim_silence(
audio: np.ndarray,
top_db: float = 25.0,
sr: int = SAMPLE_RATE,
) -> np.ndarray:
"""Trim leading/trailing silence based on energy threshold.
Args:
audio: 1D audio signal.
top_db: Threshold in dB below peak to consider as silence.
sr: Sample rate.
Returns:
Trimmed audio. Returns original if trimming would remove everything.
"""
if len(audio) == 0:
return audio
trimmed, _ = librosa.effects.trim(audio, top_db=top_db)
# Safety: don't return empty audio
if len(trimmed) < sr: # Less than 1 second
return audio
return trimmed
def preprocess_audio(audio: np.ndarray, sr: int = SAMPLE_RATE) -> np.ndarray:
"""Full preprocessing pipeline for a single audio recording.
Pipeline: bandpass → wavelet denoise → normalize → trim silence.
Args:
audio: Raw 1D audio signal.
sr: Sample rate.
Returns:
Preprocessed audio signal.
"""
audio = bandpass_filter(audio, sr=sr)
audio = wavelet_denoise(audio)
audio = normalize_amplitude(audio)
audio = trim_silence(audio, sr=sr)
return audio
def segment_audio(
audio: np.ndarray,
segment_samples: int = SEGMENT_SAMPLES,
hop_samples: int = SEGMENT_HOP,
) -> list[np.ndarray]:
"""Cut audio into fixed-length overlapping windows.
Args:
audio: 1D preprocessed audio.
segment_samples: Samples per segment (default: 3s * 2000Hz = 6000).
hop_samples: Hop between segment starts (default: 50% overlap = 3000).
Returns:
List of segments. Each is zero-padded if needed.
"""
if len(audio) == 0:
return []
segments = []
start = 0
while start < len(audio):
end = start + segment_samples
segment = audio[start:end]
# Zero-pad the last segment if shorter
if len(segment) < segment_samples:
padded = np.zeros(segment_samples, dtype=np.float32)
padded[: len(segment)] = segment
segment = padded
segments.append(segment)
start += hop_samples
return segments
def process_single_recording(
wav_path: str,
label: int,
filename: str,
output_dir: Path,
) -> list[dict]:
"""Process one recording: load → preprocess → segment → save.
Args:
wav_path: Path to .wav file.
label: Original label (-1 or 1).
filename: Recording filename (e.g., 'a0001').
output_dir: Directory to save segments.
Returns:
List of dicts with segment metadata.
"""
class_idx = LABEL_MAP[label]
audio = load_wav(wav_path)
audio = preprocess_audio(audio)
segments = segment_audio(audio)
metadata = []
for i, seg in enumerate(segments):
seg_name = f"{filename}_seg{i:04d}"
seg_path = output_dir / f"{seg_name}.npy"
np.save(str(seg_path), seg)
metadata.append({
"segment_name": seg_name,
"source_file": filename,
"segment_idx": i,
"class_idx": class_idx,
"label": label,
"segment_path": str(seg_path),
})
return metadata
def run_phase2(master_csv: Optional[str] = None) -> pd.DataFrame:
"""Execute the full Phase 2 preprocessing pipeline.
Args:
master_csv: Path to master_labels.csv. Auto-detected if None.
Returns:
DataFrame of all segment metadata.
"""
logging.basicConfig(level=logging.INFO, format="%(levelname)s: %(message)s")
logger.info("=" * 60)
logger.info("PHASE 2 — Audio Preprocessing")
logger.info("=" * 60)
# Load master labels
if master_csv is None:
master_csv = str(OUTPUT_DIR / "master_labels.csv")
master = pd.read_csv(master_csv)
logger.info("Loaded %d recordings from master labels", len(master))
# Filter to readable files only
if "readable" in master.columns:
master = master[master["readable"] == True].reset_index(drop=True)
logger.info("Readable recordings: %d", len(master))
# Create output directory
SEGMENTS_DIR.mkdir(parents=True, exist_ok=True)
all_metadata = []
total = len(master)
for idx, row in master.iterrows():
try:
meta = process_single_recording(
wav_path=row["wav_path"],
label=row["label"],
filename=row["filename"],
output_dir=SEGMENTS_DIR,
)
all_metadata.extend(meta)
except Exception as e:
logger.warning("Error processing %s: %s", row["filename"], e)
if (idx + 1) % 200 == 0 or idx == total - 1:
logger.info("Processed %d/%d recordings → %d segments so far",
idx + 1, total, len(all_metadata))
# Save segment metadata
seg_df = pd.DataFrame(all_metadata)
seg_csv = OUTPUT_DIR / "segments_metadata.csv"
seg_df.to_csv(seg_csv, index=False)
logger.info("Saved %d segments metadata to %s", len(seg_df), seg_csv)
# Print stats
normal = (seg_df["class_idx"] == 0).sum()
abnormal = (seg_df["class_idx"] == 1).sum()
logger.info("Segments: %d normal / %d abnormal / %d total",
normal, abnormal, len(seg_df))
return seg_df
if __name__ == "__main__":
run_phase2()
|