1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
|
import numpy
import numpy as np
from scipy.signal import firwin, lfilter, decimate
import soundfile
def load_pdm_file(filename):
with open(filename, 'rb') as f:
byte_data = np.frombuffer(f.read(), dtype=np.uint8)
bit_data = np.unpackbits(byte_data)
# Convert to +1 (for 1s) and -1 (for 0s)
#pdm_signal = 2 * bit_data - 1
pdm_signal = bit_data
return pdm_signal
class CICFilter:
def __init__(self, decimation=64, delay=1, stages=3):
self.R = decimation
self.M = delay
self.N = stages
# Integrator and comb delay lines
self.integrators = [0] * self.N
self.comb_buffers = [[0] * self.M for _ in range(self.N)]
self.comb_indices = [0] * self.N
self.input_count = 0
def process_sample(self, sample):
sample = int(sample) # ensure it's Python int
for i in range(self.N):
self.integrators[i] = int(self.integrators[i] + sample)
sample = self.integrators[i]
self.input_count += 1
if self.input_count == self.R:
self.input_count = 0
comb_input = sample
for i in range(self.N):
idx = self.comb_indices[i]
delayed = self.comb_buffers[i][idx]
self.comb_buffers[i][idx] = comb_input
self.comb_indices[i] = (idx + 1) % self.M
comb_input = int(comb_input - delayed)
return comb_input
else:
return None
def process(self, signal):
output = []
for s in signal:
y = self.process_sample(s)
if y is not None:
output.append(y)
return output
def pdm_to_pcm(pdm_signal, decimation_factor=64, filter_type='fir', filter_kwargs={}):
if filter_type == 'fir':
# Design a low-pass filter
defaults = dict(numtaps=501, cutoff=0.5/decimation_factor)
kwargs = {}
kwargs.update(defaults)
kwargs.update(filter_kwargs)
fir_filter = firwin(**kwargs)
# Filter the PDM signal
filtered = lfilter(fir_filter, [1.0], pdm_signal)
# Decimate (downsample)
pcm_signal = filtered[::decimation_factor]
elif filter_type == 'cic':
# Use CIC filter
defaults = dict(stages=3, delay=1)
kwargs = {}
kwargs.update(defaults)
kwargs.update(filter_kwargs)
pdm_signal = pdm_signal.astype(int)
cic = CICFilter(decimation=decimation_factor, **kwargs)
# Convert 1-bit PDM to +/-1 format
pdm_signal = [2*s - 1 for s in pdm_signal.astype(int)]
pcm_signal = cic.process(pdm_signal)
pcm_signal = numpy.array(pcm_signal, dtype=float)
return pcm_signal
def parse():
import argparse
parser = argparse.ArgumentParser(description='Process an input file and write to an output file.')
parser.add_argument('-i', '--input', type=str, required=True, help='Path to the input file')
parser.add_argument('-o', '--output', type=str, required=True, help='Path to the output file')
parser.add_argument('--filter', type=str, default='fir')
parser.add_argument('--samplerate', type=int, default=16000)
parser.add_argument('--oversample', type=int, default=64)
args = parser.parse_args()
return args
def main():
args = parse()
pdm_path = args.input
out_path = args.output
pdm_data = load_pdm_file(pdm_path)
print('loaded', pdm_data.shape)
# Convert to PCM
oversample = args.oversample
samplerate = args.samplerate
pcm_data = pdm_to_pcm(pdm_data, decimation_factor=oversample, filter_type=args.filter)
# Normalize and save to WAV
pcm_data = np.expand_dims(pcm_data, axis=1)
#print(pcm_data.shape)
pcm_data -= numpy.mean(pcm_data) # remove DC offset
pcm_data = pcm_data / (numpy.max(numpy.abs(pcm_data)) * 2.0) # normalize scale
print(numpy.min(pcm_data), numpy.max(pcm_data), numpy.mean(pcm_data), pcm_data.dtype)
soundfile.write(out_path, pcm_data, samplerate, subtype='PCM_16')
print('Wrote', out_path)
if __name__ == '__main__':
main()
|
