CDイメージからPCMを抽出してMS分離の有無でADPCMを適用したときの歪みを比較する

ms_codec.py

import numpy as np
from scipy import signal
from scipy import fftpack
import matplotlib
matplotlib.use('Agg')
import matplotlib.pyplot as plt
import soundfile as sf
import audioop

samplerate = 44100
fft_size = 8192

start_time = [15, 30, 7]  # [min, sec, 0.01sec]
stop_time = [18, 54, 67]
start_sec = (start_time[0] * 60) + start_time[1] + (start_time[2] * 0.01)
stop_sec = (stop_time[0] * 60) + stop_time[1] + (stop_time[2] * 0.01)
start_sample = int(start_sec * samplerate)
stop_sample = int(stop_sec * samplerate)

num_sample = stop_sample - start_sample
num_seek = start_sample * 2 * 2  # 16bit, 2ch

time_min = np.linspace(0., (stop_sec-start_sec)/60, num_sample, endpoint=False)

file = open("Image.bin", mode="rb")
file.seek(num_seek)

data = np.fromfile(file, "<i2", num_sample * 2)  # 16bit, 2ch
data = data.reshape((-1,2))

data_L = data[:,0].astype(np.int32)
data_R = data[:,1].astype(np.int32)
del(data)

data_mid = ((data_L + data_R) / 2).astype(np.int32)
data_side = ((data_L - data_R) / 2).astype(np.int32)

def adopt_codec(data):
    inpcm = data.astype(np.int16).tobytes()
    adpcm_frag, _ = audioop.lin2adpcm(inpcm, 2, None)
    outpcm, _ = audioop.adpcm2lin(adpcm_frag, 2, None)
    return np.frombuffer(outpcm, dtype=np.int16)

data_L_cod = adopt_codec(data_L)[:num_sample-4]
data_R_cod = adopt_codec(data_R)[:num_sample-4]
data_mid_cod = adopt_codec(data_mid)[:num_sample-4]
data_side_cod = adopt_codec(data_side)[:num_sample-4]
data_L_frm_ms = (data_mid_cod + data_side_cod).astype(np.int16)
data_R_frm_ms = (data_mid_cod - data_side_cod).astype(np.int16)

dist_LR = np.sum(np.power(np.concatenate([data_L_cod, data_R_cod]) - np.concatenate([data_L[:num_sample-4], data_R[:num_sample-4]]), 2)) / num_sample
dist_msLR = np.sum(np.power(np.concatenate([data_L_frm_ms, data_R_frm_ms]) - np.concatenate([data_L[:num_sample-4], data_R[:num_sample-4]]), 2)) / num_sample

print('LR_distortion: {}'.format(dist_LR))
print('MS_distortion: {}'.format(dist_msLR))

sf.write("outdata_LR.wav", np.array([data_L_cod, data_R_cod]).T.astype(np.int16), samplerate)
sf.write("outdata_MS.wav", np.array([data_L_frm_ms, data_R_frm_ms]).T.astype(np.int16), samplerate)