how to calculate signal to noise ratio using python

Admin 1
Viewed 28477

Dear experts i have a data set.i just want to calculate signal to noise ratio of the data. data is loaded here https://i.fluffy.cc/jwg9d7nRNDFqdzvg1Qthc0J7CNtKd5CV.html

my code is given below:

import numpy as np
from scipy import signaltonoise
import scipy.io
dat=scipy.io.loadmat('./data.mat')
arr=dat['dn']
snr=scipy.stats.signaltonoise(arr, axis=0, ddof=0)

but i am getting error like importError: cannot import name 'signaltonoise' from 'scipy' if it doesn't exist how to calculate snr,please suggest some other way to do it with this data set using python.

scipy signal-processing python
3 Answers

scipy.stats.signaltonoise was removed in scipy 1.0.0. You can either downgrade your scipy version or create the function yourself:

def signaltonoise(a, axis=0, ddof=0):
    a = np.asanyarray(a)
    m = a.mean(axis)
    sd = a.std(axis=axis, ddof=ddof)
    return np.where(sd == 0, 0, m/sd)

Source: https://github.com/scipy/scipy/blob/v0.16.0/scipy/stats/stats.py#L1963

See the github link for the docstring.

Edit: the full script would then look as follows

import numpy as np
import scipy.io

def signaltonoise(a, axis=0, ddof=0):
    a = np.asanyarray(a)
    m = a.mean(axis)
    sd = a.std(axis=axis, ddof=ddof)
    return np.where(sd == 0, 0, m/sd)

dat = scipy.io.loadmat('./data.mat')
arr = dat['dn']
snr = signaltonoise(arr)
user_59040590

More generally speaking, it depends on the application. For many applications, the relation between mean and standard deviation might be sufficient.

As JuliettVictor pointed out, the old scipy implementation's source code can be found online easily and is the most common one. In order to convert this to decibels, one would add . Before that one should calculate the absolute value, in case the signal's mean is negative:

def signaltonoise_dB(a, axis=0, ddof=0):
    a = np.asanyarray(a)
    m = a.mean(axis)
    sd = a.std(axis=axis, ddof=ddof)
    return 20*np.log10(abs(np.where(sd == 0, 0, m/sd)))

This runs into problems though, when the signal of interest contains higher frequencies (e.g. in audio applications, here, DC is often filtered out even).

In matlab's snr() function a kaiser windowed periodogram is used, the peak of the fundamental is detected and its power is computed. The same happens for the harmonics. The rest of the signal is assumed to be noise and their corresponding power levels are calculated. Other approaches involve low-pass filtering of the signal (similar to calculating its mean).
Yet another python based example can be found here. An incomplete overview of methods (including a matlab like periodogram-based one) in python can be found here

user_54772340

You can use Sox for that:

def add_noise(exp,sig, n_directory, snrdb):
    """     This function calculate the mixture of an audio 
        and an audio file as a noise with a desired SNR in db
Input:
    sig: signal, the output vector of wav.read
    rate: the signal's rate
    n_directory: directory of noise
    snrdb: Desired signal-to-noise ratio in db

Output:
    d = s + n such that SNR = Ps/Pn     """
direc_st = exp + "mixed.wav"

write(exp + 'sig_t.wav', 16000, sig)
s = np.expand_dims(sig, -1)
s = s.astype('float64')

rate, n = wav.read(n_directory)
n = np.expand_dims(n, -1)
n = n.astype('float64')

snr = 10 ** (snrdb * 0.1)
Es = np.sum(s ** 2)
En = np.sum(n ** 2)

snr = 10 ** (snrdb * 0.1)
alpha = np.sqrt(Es / (snr * En))

cmd = "sox -m -v" + str(
    alpha) + " " + n_directory + ' ' + exp + 'sig_t.wav ' + direc_st
os.system(cmd)

rate, d = wav.read(direc_st)

return d
user_143049540
Related