CODEWITHSUNDEEP
pythonmatlabsignal-processing
Anand PA
Anand PA

Reputation: 125

Python equivalent of Matlab's resample()

Is there a python function that achieves resampling in the way MATLAB's resample() does? I've looked into scikits.samplerate's resample function but I'm not quite getting similar results.

Upvotes: 2

Views: 6148

Answers (4)

Masoud Nateghi
Masoud Nateghi

Reputation: 1

I came across the same problem and developed this wrapper function, which performs exactly the same as the resample function in MATLAB.

def resample(x, p, q, n=10, beta=5):
    from scipy.signal import upfirdn, lfilter, firwin
    # anti-aliasing filter design
    maxpq = max(p, q)
    fc = 1 / maxpq
    order = 2 * n * maxpq
    b = firwin(order + 1, fc, window=('kaiser', beta))
    b = p * b / np.sum(b)

    # upsampling
    x_us = upfirdn([1], x, p)

    # anti-aliasing filtering
    x_aa = lfilter(b, 1, np.hstack((x_us, np.zeros(q * n))))

    # downsampling
    x_ds = upfirdn([1], x_aa, 1, q)
    x_ds = x_ds[n:]

    return x_ds

Upvotes: 0

Brad123
Brad123

Reputation: 944

After looking at this for too long here's the best equivalent to matlab's resample(x,p,q):

x_resampled = resampy.resample(x, q, p, filter='sinc_window', window=scipy.signal.flattop)

For variance compared to the MATLAB results I'm seeing, 96.59 (length > 5000). As a comparison when I use scipy's 'resample',

resample_len = int(len(x) * p / q)+1    
x_resampled  = resample(x, resample_len , t=None, axis=0, window=None, domain='time')

I get a variance of 4,281.

Exploring the MATLAB method more, here's the description:

y = resample(x,p,q) resamples the input sequence, x, at p/q times the original sample rate. resample applies an FIR Antialiasing Lowpass Filter to x and compensates for the delay introduced by the filter. The function operates along the first array dimension with size greater than 1.

Where the FIR filter is:

p = 3;
q = 2;
maxpq = max(p,q);
fc = 1/maxpq;
n = 10;
order = 2*n*maxpq;
beta = 5;
b = fir1(order,fc,kaiser(order+1,beta));
b = p*b/sum(b);

I didn't discover a true 1-to-1 replacement, nor did I find one looking online or in ChatGPT.

Applying all the windows using window=scipy.signal.some_window, here's the name and variance:

boxcar          150.06
triang          102.92
blackman        107.66
hamming         117.09
hann            115.33
bartlett        102.91
flattop         96.59
parzen          103.51
bohman          106.08
blackmanharris  102.41
nuttall         102.71
barthann        111.97
cosine          127.05
exponential     13535.85
tukey           144.04

As you can see flattop performed the closest. If someone out there can get a better correlation to MATLAB's method please post it.

Upvotes: 0

Ohads
Ohads

Reputation: 63

The best function that is equivalent to MATLAB resample is as such:

MATLAB:

resample(  Data_Low_Freq_v, Upsample_n ,Downsample_n );

Python:

import scipy.signal as ssg    
ssg.resample_poly(Data_Low_Freq_v, Upsample_n, Downsample_n)

Upvotes: -1

Mike Tang
Mike Tang

Reputation: 46

There's a blog 'Audio Resampling in Python' about this. Both resampy and scipy.signal have a resample function implementation in different ways.

Upvotes: 3

Related Questions