edgevoid
68db84a10c
- 两个虚拟声卡: 5.1 直通版 / 立体声矩阵上混版 - 全链路 96kHz/24bit, SADIE-II 真人 HRTF 96K/512tap - 命令: 空间音频(开关) / mpv-影院(单程序) - deb 打包 + 验收测试脚本 - README 记录 9 个实测踩坑(增益校准/采样率/相位等)
70 lines
2.5 KiB
Bash
70 lines
2.5 KiB
Bash
#!/bin/bash
|
|
# 响度校准: HRTF 卷积会把能量摊薄, 算准该补多少 dB 再补, 别拍脑袋
|
|
set -eu
|
|
cd "/home/lou/桌面/工作区/实验/空间音频"
|
|
SOFA=/usr/share/libmysofa/MIT_KEMAR_normal_pinna.sofa
|
|
|
|
echo "===== 1. 量现有两版的响度 ====="
|
|
python3 - <<'PY'
|
|
import math
|
|
import struct
|
|
import wave
|
|
|
|
|
|
def stat(path: str) -> tuple[float, float]:
|
|
"""返回 (峰值, 均方根), 都归一化到 1.0。"""
|
|
with wave.open(path) as w:
|
|
channels, frames = w.getnchannels(), w.getnframes()
|
|
raw = w.readframes(frames)
|
|
data = struct.unpack("<%dh" % (len(raw) // 2), raw)
|
|
peak = max(abs(x) for x in data) / 32768.0
|
|
rms = (sum(x * x for x in data) / len(data)) ** 0.5 / 32768.0
|
|
return peak, rms
|
|
|
|
|
|
peak_a, rms_a = stat("A-普通下混.wav")
|
|
peak_b, rms_b = stat("B-HRTF环绕.wav")
|
|
need = 20 * math.log10(rms_a / rms_b)
|
|
print(" A 普通下混 峰值 %.3f (%.1f dBFS) 均方根 %.4f (%.1f dBFS)"
|
|
% (peak_a, 20 * math.log10(peak_a), rms_a, 20 * math.log10(rms_a)))
|
|
print(" B HRTF 峰值 %.3f (%.1f dBFS) 均方根 %.4f (%.1f dBFS)"
|
|
% (peak_b, 20 * math.log10(peak_b), rms_b, 20 * math.log10(rms_b)))
|
|
print(" => 需补 %.1f dB; 补后 B 峰值约 %.1f dBFS" % (need, 20 * math.log10(peak_b) + need))
|
|
PY
|
|
|
|
echo
|
|
echo "===== 2. 按算出的量重渲染 B 版 ====="
|
|
GAIN=$(python3 -c "
|
|
import math, struct, wave
|
|
def rms(p):
|
|
with wave.open(p) as w:
|
|
c, n = w.getnchannels(), w.getnframes()
|
|
raw = w.readframes(n)
|
|
d = struct.unpack('<%dh' % (len(raw)//2), raw)
|
|
return (sum(x*x for x in d)/len(d))**0.5
|
|
print('%.1f' % (20*math.log10(rms('A-普通下混.wav')/rms('B-HRTF环绕.wav'))))
|
|
")
|
|
echo " gain = ${GAIN} dB"
|
|
ffmpeg -y -hide_banner -loglevel warning -i 51-test.wav \
|
|
-af "sofalizer=sofa=$SOFA:type=freq:radius=1.2:gain=${GAIN}" \
|
|
-c:a pcm_s16le "B-HRTF环绕.wav"
|
|
|
|
echo
|
|
echo "===== 3. 复测(峰值别顶到 0dBFS, 否则削波) ====="
|
|
python3 - <<'PY'
|
|
import math
|
|
import struct
|
|
import wave
|
|
|
|
for f in ("A-普通下混.wav", "B-HRTF环绕.wav"):
|
|
with wave.open(f) as w:
|
|
channels, frames = w.getnchannels(), w.getnframes()
|
|
raw = w.readframes(frames)
|
|
data = struct.unpack("<%dh" % (len(raw) // 2), raw)
|
|
peak = max(abs(x) for x in data) / 32768.0
|
|
rms = (sum(x * x for x in data) / len(data)) ** 0.5 / 32768.0
|
|
clip = sum(1 for x in data if abs(x) >= 32767)
|
|
print(" %-18s 峰值 %.1f dBFS 均方根 %.1f dBFS 削波点 %d"
|
|
% (f, 20 * math.log10(peak), 20 * math.log10(rms), clip))
|
|
PY
|