chore: remove unused imports and dead code

sync_audio.py

@@ -15,7 +15,6 @@ Usage:
 from __future__ import annotations
 import os
 import sys
-import math
 from typing import Tuple, Literal
 
 import numpy as np
@@ -120,24 +119,14 @@ def correlate(
             # fftshift to center 0 lag
             corr_t = torch.roll(corr_t, shifts=N // 2)
             corr = corr_t.cpu().numpy()
-            # keep only valid length n centered
-            start = N // 2 - (len(b) - 1)
-            # Actually after roll, we want slice; simpler: use raw without roll and map lag later
-            # Recompute without roll for correct indexing, keep roll for display but fix lag calc
-            # For lag calc we need not shift: peak at N - lag etc. So we use shifted version + offset mapping
-            # To avoid confusion, return shifted corr and map lag accordingly
-            peak_idx = int(np.argmax(np.abs(corr)))
-            # lag = peak_idx - (len(b)-1) - (N//2 - (len(b)-1))? Let's instead use unshifted for lag and shifted for display
-            # Fallback: compute lag from unshifted
+            # compute lag from unshifted correlation for correct indexing
             corr_raw = torch.fft.irfft(R, n=N).cpu().numpy()
             peak_raw = int(np.argmax(np.abs(corr_raw)))
-            # raw layout: lag 0 at 0, positive lag (b delayed) wraps to end
             if peak_raw > N // 2:
                 lag = peak_raw - N
             else:
                 lag = peak_raw
             peak_val = float(np.abs(corr_raw[peak_raw]))
-            # return shifted display corr + lag
             return lag, peak_val, corr
         except Exception as e:
             print(f"torch cuda failed ({e}), falling back cpu", file=sys.stderr)
@@ -156,10 +145,7 @@ def correlate(
     else:
         lag = peak_raw
     peak_val = float(np.abs(corr_raw[peak_raw]))
-    # shifted for display centered at 0
     corr_shifted = np.fft.fftshift(corr_raw)
-    # but we return raw shifted centered? Provide shifted for plotting
-    # For consistency with torch branch we return shifted display array
     return lag, peak_val, corr_shifted
 
 
@@ -170,9 +156,7 @@ def lag_to_seconds(lag_samples: int, sr: int = 48000) -> float:
 def apply_lag_with_silence(wav: np.ndarray, lag_samples: int) -> np.ndarray:
     """
     Apply lag by padding silence at start (no sound for delay).
-    lag >0: b delayed vs a => pad b at start with lag silence to align to a, or trim a.
-    For mux: if lag>0 (b starts later), we need to delay a by lag or提前 b.
-    This helper pads with zeros (silence) at start; negative lag pads the other direction.
+    lag > 0: b delayed vs a. lag < 0: a delayed vs b.
     Caller decides which file to pad.
     """
     if lag_samples == 0:
@@ -200,7 +184,7 @@ def device_info() -> str:
 
 
 if __name__ == "__main__":
-    import argparse, pathlib
+    import argparse
     ap = argparse.ArgumentParser()
     ap.add_argument("a", help="wav A")
     ap.add_argument("b", help="wav B")