Example showing processing time difference for filtering numpy signals along different axes.

scipy_filt_axes.py

import numpy as np
from scipy import signal

fs = 30000  # Hz
n_chans = 100
sig_dur = 300.0  # seconds
lp_cutoff = 250  # Hz
lp_order = 8
sig_freq_low = 14  # Frequency of sine wave in signal
sig_amp_low = 10.0  # Amplitude of sine wave in signal
sig_freq_high = 500
sig_amp_high = 5.0

t = np.linspace(0, sig_dur, fs)
sig = sig_amp_low * np.sin(2 * np.pi * sig_freq_low * t) + sig_amp_high * np.sin(2 * np.pi * sig_freq_high * t)
# Create matrices. Use broadcast addition.
# Note that we cannot simply use the transpose because .T does not change the memory layout.
sig_R = sig[:, None] + np.zeros((len(t), n_chans))  # time in Rows
sig_C = sig[None, :] + np.zeros((n_chans, len(t)))  # time in Columns
print("Is c_contiguous? sig_R: ", sig_R.shape, sig_R.flags.c_contiguous,
      "; sig_C: ", sig_C.shape, sig_C.flags.c_contiguous)
# Create the filter
sos = signal.butter(lp_order, lp_cutoff/(fs/2), output='sos')
# Time signal filtering
%timeit signal.sosfiltfilt(sos, sig_R, axis=0)
# 485 ms ± 33.7 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)
%timeit signal.sosfiltfilt(sos, sig_C, axis=-1)
# 235 ms ± 12.1 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)

# Optional, plot to verify the filter operation works as intended.
import matplotlib.pyplot as plt
plt.figure()
plt.plot(t, sig)
# plot filtered signals with offsets so you can actually see them.
plt.plot(t, signal.sosfiltfilt(sos, sig_R, axis=0)[:, 0] - 1)  # orange
plt.plot(t, signal.sosfiltfilt(sos, sig_C, axis=-1)[0, :] + 1)  # green
plt.show()