modified parameters

bridge.py

@@ -6,10 +6,10 @@ import struct
 import numpy as np
 import fcntl
 
-# Lobotomize the Linux kernel pipe so it can't hoard future audio
+# Loosen the lobotomy slightly to 8192 bytes. 
 try:
     F_SETPIPE_SZ = 1031
-    fcntl.fcntl(sys.stdin.fileno(), F_SETPIPE_SZ, 4096)
+    fcntl.fcntl(sys.stdin.fileno(), F_SETPIPE_SZ, 8192)
 except Exception:
     pass
 
@@ -17,22 +17,22 @@ except Exception:
 UDP_IP = os.environ.get("UDP_IP", "239.0.0.1")
 UDP_PORT = int(os.environ.get("UDP_PORT", "11988"))
 
-# Audio stream constants
-SAMPLE_RATE = 44100
-CHUNK_SIZE = 1024
+# Audio stream constants optimized for 24kHz downsampling
+SAMPLE_RATE = 24000
+CHUNK_SIZE = 512
 CHUNK_DURATION = CHUNK_SIZE / SAMPLE_RATE
 FRAME_BYTES = CHUNK_SIZE * 4
-MAX_SILENT_FRAMES = 43  # ~1 second of silence at 43.06 FPS
+MAX_SILENT_FRAMES = 43
 
-# Hardcoded DSP tuning
-FREQ_MIN = 44.0          # Aligned with physical 43.06Hz FFT bin resolution
-FREQ_MAX = 12000.0       # Capture crisp cymbals and high transients
+# Hardcoded DSP tuning for violent clipping
+FREQ_MIN = 44.0
+FREQ_MAX = 12000.0
 SILENCE_THRESHOLD = 0.5
-GAIN_MIN = 800.0         # Floor: prevents squashing heavily mastered EDM
-GAIN_MAX = 8500.0        # Ceiling: prevents boosting background tape hiss
-TILT_EXPONENT = 0.42     # Treble tilt compensation
-DECAY_RATE = 0.95        # Faster recovery (~1s) for actual snap and bounce
-CONTRAST_EXPONENT = 1.2  # >1.0 crushes noise floor & exaggerates beats
+GAIN_MIN = 800.0
+GAIN_MAX = 1500.0
+TILT_EXPONENT = 0.45
+DECAY_RATE = 0.40
+CONTRAST_EXPONENT = 4.5
 
 sock = socket.socket(socket.AF_INET, socket.SOCK_DGRAM, socket.IPPROTO_UDP)
 sock.setsockopt(socket.IPPROTO_IP, socket.IP_MULTICAST_TTL, 2)
@@ -49,7 +49,6 @@ band_centers = np.sqrt(FREQ_EDGES[:-1] * FREQ_EDGES[1:])
 TILT_WEIGHTS = ((band_centers / FREQ_MIN) ** TILT_EXPONENT).astype(np.float32)
 
 # Build a C-optimized Matrix for dot-product binning
-# Shape: (16 bands, 513 FFT bins). 
 BIN_MATRIX = np.zeros((16, len(fft_freqs)), dtype=np.float32)
 
 for i in range(16):
@@ -62,7 +61,6 @@ for i in range(16):
         closest = non_zero_bins[np.argmin(np.abs(fft_freqs[non_zero_bins] - low))]
         idx = [closest]
         
-    # The mean is just 1.0 / count. Multiply by the tilt weight immediately.
     BIN_MATRIX[i, idx] = (1.0 / len(idx)) * TILT_WEIGHTS[i]
 
 # State variables
@@ -71,21 +69,28 @@ silence_frames = 0
 running_peak = 0.05
 STRUCT_FMT_V2 = "<6s2xffB3x16sd"
 
+def read_exact_chunk(fd, size):
+    buf = bytearray(size)
+    view = memoryview(buf)
+    pos = 0
+    while pos < size:
+        chunk = fd.readinto(view[pos:])
+        if not chunk:
+            return None
+        pos += chunk
+    return bytes(buf)
+
 start_time = None
 frames_processed = 0
 
 while True:
     t0 = time.perf_counter()
-    raw_data = sys.stdin.buffer.read(FRAME_BYTES)
+    raw_data = read_exact_chunk(sys.stdin.buffer, FRAME_BYTES)
     read_duration = time.perf_counter() - t0
 
-    # Short-read survival patch
     if not raw_data:
         break
-    if len(raw_data) < FRAME_BYTES:
-        continue
 
-    # Micro-reset baseline to permanently kill accumulated audio pipeline drift
     if read_duration > 0.015:
         start_time = time.perf_counter()
         frames_processed = 0
@@ -94,7 +99,6 @@ while True:
     if start_time is None:
         start_time = time.perf_counter()
 
-    # Audio ingestion and magnitude extraction
     audio = np.frombuffer(raw_data, dtype=np.int16).astype(np.float32)
     left = audio[0::2]
     right = audio[1::2]
@@ -113,20 +117,16 @@ while True:
         windowed = mono * HANNING_WINDOW
         fft_vals = np.abs(np.fft.rfft(windowed)) * (2.0 / CHUNK_SIZE)
 
-        # C-optimized single matrix multiplication replaces the 16-step for loop and tilt math
         tilted_energies = BIN_MATRIX @ fft_vals
 
-        # Track rolling peak
         current_max = float(np.max(tilted_energies))
         if current_max > running_peak:
             running_peak = current_max
         else:
             running_peak = max(0.005, running_peak * DECAY_RATE)
 
-        # Dynamic gain bounded by floor and ceiling
         dynamic_gain = np.clip(1.0 / running_peak, GAIN_MIN / 255.0, GAIN_MAX / 255.0)
 
-        # Normalize 0.0 to 1.0, apply contrast exponent, scale to 0-255
         normalized = np.clip(tilted_energies * dynamic_gain, 0.0, 1.0)
         contrasted = (normalized ** CONTRAST_EXPONENT) * 255.0
         fft_result = bytes(np.clip(contrasted, 0, 255).astype(np.uint8))
@@ -146,13 +146,17 @@ while True:
         except Exception:
             pass
 
-    # Metronome pacing
+    # Hybrid PLL Metronome to provide flawless backpressure
     frames_processed += 1
     target_time = start_time + (frames_processed * CHUNK_DURATION)
     sleep_time = target_time - time.perf_counter()
 
-    if sleep_time > 0:
-        time.sleep(sleep_time)
-    elif sleep_time < -1.0:
+    if sleep_time > 0.002:
+        time.sleep(sleep_time - 0.001)
+
+    while time.perf_counter() < target_time:
+        pass
+
+    if sleep_time < -1.0:
         start_time = time.perf_counter()
         frames_processed = 0

docker-compose.yml

@@ -13,7 +13,6 @@ services:
       $${LMS_IP:+-s $$LMS_IP}
       -n "$${PLAYER_NAME:-WLED-Audio-Sync}"
       -m "$${PLAYER_MAC:-02:00:00:11:98:88}"
-      -b 2000:20
       -o -
       -r 44100
       -d all=info | python3 bridge.py'