e75c4d6f07
- Remove session-ses_2f27.md (161KB raw session log) - Remove 49 ROOT_* duplicate files across REFERENCE/ - Remove 14 duplicate files between REFERENCE/ root and history/ - Remove asr_legacy.rs (dead code, replaced by asr.rs) - Remove src/core/worker/ (duplicate JobWorker) - Remove src/core/layers/ (empty directory) - Remove 4 .bak files in src/ - Remove 7 dead private methods in worker/processor.rs - Remove backup directory from git tracking
126 lines
4.3 KiB
Python
126 lines
4.3 KiB
Python
#!/opt/homebrew/bin/python3.11
|
|
"""
|
|
Sound Event Detector (Impulse/Gunshot)
|
|
職責：使用聲學特徵檢測高能量脈衝聲音 (如槍聲、爆炸)。
|
|
"""
|
|
|
|
import librosa
|
|
import numpy as np
|
|
import json
|
|
import os
|
|
import sys
|
|
|
|
# 設定
|
|
OUTPUT_DIR = os.getenv("MOMENTRY_OUTPUT_DIR", "./output")
|
|
UUID = os.getenv("UUID", "384b0ff44aaaa1f1")
|
|
AUDIO_PATH = os.path.join(OUTPUT_DIR, UUID, f"{UUID}.wav")
|
|
OUTPUT_JSON = os.path.join(OUTPUT_DIR, UUID, f"{UUID}.sound_events.json")
|
|
|
|
|
|
def detect_impulse_sounds(audio_path, threshold_multiplier=1.5):
|
|
"""
|
|
檢測脈衝聲音 (Impulse Sounds)
|
|
原理：尋找 RMS 能量的局部峰值，且該峰值顯著高於背景噪音。
|
|
"""
|
|
print(f"🔊 Loading audio: {audio_path}")
|
|
# 載入音頻 (Mono, 22050Hz)
|
|
y, sr = librosa.load(audio_path, sr=22050)
|
|
|
|
print("📊 Analyzing energy envelope...")
|
|
# 1. 計算 RMS 能量 (以 0.05秒 為一幀)
|
|
frame_length = int(0.05 * sr)
|
|
hop_length = int(0.02 * sr)
|
|
rms = librosa.feature.rms(y=y, frame_length=frame_length, hop_length=hop_length)[0]
|
|
|
|
# 2. 計算動態閾值 (背景噪音 + 標準差的倍數)
|
|
# 使用移動平均來適應不同場景的背景音
|
|
background = np.median(rms)
|
|
threshold = background * threshold_multiplier + 0.05 # 絕對底限
|
|
|
|
print(f" Background Level: {background:.4f}")
|
|
print(f" Detection Threshold: {threshold:.4f}")
|
|
|
|
# 3. 尋找超過閾值的峰值
|
|
# 使用 scipy 的 find_peaks 或簡單的 numpy 邏輯
|
|
from scipy.signal import find_peaks
|
|
|
|
peaks, properties = find_peaks(
|
|
rms, height=threshold, distance=int(0.2 / 0.02)
|
|
) # 至少間隔 0.2秒
|
|
|
|
# 4. 過濾與分類
|
|
events = []
|
|
for peak_idx in peaks:
|
|
# 時間戳 (秒)
|
|
time_sec = peak_idx * hop_length / sr
|
|
|
|
# 特徵分析：檢查頻譜質心 (Spectral Centroid) - 槍聲通常頻譜質心高
|
|
# 取峰值前後一小段
|
|
start_frame = max(0, peak_idx - 2)
|
|
end_frame = min(len(rms), peak_idx + 2)
|
|
frame_idx = int(time_sec * sr)
|
|
segment = y[max(0, frame_idx - 1000) : frame_idx + 1000]
|
|
|
|
if len(segment) > 0:
|
|
# 計算頻譜質心 (聲音的 "亮度")
|
|
centroid = librosa.feature.spectral_centroid(y=segment, sr=sr)[0]
|
|
avg_centroid = np.mean(centroid)
|
|
|
|
# 計算頻帶能量 (Gunshot 通常高頻能量豐富)
|
|
# 這裡簡化：如果 RMS 極高，直接標記為 "Gunshot/Explosion"
|
|
rms_val = rms[peak_idx]
|
|
|
|
event_type = "Loud Noise"
|
|
if rms_val > threshold * 2.0:
|
|
event_type = "Explosion/Gunshot" # 極高能量
|
|
elif rms_val > threshold * 1.2:
|
|
event_type = "Loud Impact"
|
|
|
|
events.append(
|
|
{
|
|
"timestamp": round(time_sec, 2),
|
|
"type": event_type,
|
|
"energy": round(float(rms_val), 4),
|
|
"centroid": round(float(avg_centroid), 2),
|
|
}
|
|
)
|
|
|
|
return events
|
|
|
|
|
|
if __name__ == "__main__":
|
|
if not os.path.exists(AUDIO_PATH):
|
|
# 嘗試從 mp4 提取
|
|
AUDIO_PATH_MP4 = os.path.join(OUTPUT_DIR, UUID, f"{UUID}.mp4")
|
|
if not os.path.exists(AUDIO_PATH_MP4):
|
|
AUDIO_PATH_MP4 = os.path.join(OUTPUT_DIR, UUID, f"{UUID}.mov")
|
|
|
|
if os.path.exists(AUDIO_PATH_MP4):
|
|
print("🎥 Extracting audio from video...")
|
|
os.system(f"ffmpeg -y -i {AUDIO_PATH_MP4} -vn -ar 16000 -ac 1 {AUDIO_PATH}")
|
|
else:
|
|
print("❌ No audio/video found.")
|
|
sys.exit(1)
|
|
|
|
print(f"🕵️♂️ Starting Sound Event Detection for {UUID}...")
|
|
|
|
# 執行檢測
|
|
events = detect_impulse_sounds(AUDIO_PATH)
|
|
|
|
# 保存結果
|
|
with open(OUTPUT_JSON, "w") as f:
|
|
json.dump({"sound_events": events}, f, indent=2)
|
|
|
|
print(f"\n🎉 Found {len(events)} potential sound events.")
|
|
print(f"💾 Results saved to {OUTPUT_JSON}")
|
|
|
|
# 顯示前 10 個高能量事件
|
|
print("\n🔥 Top 10 Loudest Events (Potential Gunshots):")
|
|
# 按能量排序
|
|
sorted_events = sorted(events, key=lambda x: x["energy"], reverse=True)[:10]
|
|
for i, ev in enumerate(sorted_events):
|
|
m, s = divmod(ev["timestamp"], 60)
|
|
print(
|
|
f" {i + 1}. [{int(m):02d}:{s:05.2f}] {ev['type']} (Energy: {ev['energy']:.4f})"
|
|
)
|