#!/usr/bin/env python3.11
#
# A simple demonstration on how to compute signal-to-noise ratio of an audio (wav) file
# Note that computing the SNR of an audio is a non-trivial task, since it is not clear how to separate a generic audio into the "signal" and "noise".
# The solution demonstrated here applies a simple bandpass filter to extract the "signal", i.e., within a frequency range of interest. Frequencies outside of the band are considered noise.
#
# Stilianos Louca
# Copyright 2024
#
# LICENSE AGREEMENT
# - - - - - - - - -
# All rights reserved.
# Use and redistributions of this code is permitted for commercial and non-commercial purposes,
# under the following conditions:
#
#	* Redistributions must retain the above copyright notice, this list of 
#	  conditions and the following disclaimer in the code itself, as well 
#	  as in documentation and/or other materials provided with the code.
#	* Neither the name of the original author (Stilianos Louca), nor the names 
#	  of its contributors may be used to endorse or promote products derived 
#	  from this code without specific prior written permission.
#	* Proper attribution must be given to the original author, including a 
#     reference to any peer-reviewed publication through which the code was published.
#
# THIS CODE IS PROVIDED BY THE COPYRIGHT HOLDER AND CONTRIBUTORS "AS IS" AND ANY 
# EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 
# OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 
# IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, 
# INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, 
# PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 
# HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 
# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS CODE, 
# EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
# - - - - - - - - -


import librosa
import numpy
import glob
import os

import numpy as np
import scipy.signal
import soundfile


RESAMPLING_RATE = 22050

wav_filepaths = glob.glob("input/number_individual_recordings/**/*.wav", recursive=True)
print("Found %d wav files"%(len(wav_filepaths)))

# function for computing the signal-to-noise ratio of an audio (basically a 1D numpy array)
# The SNR is the ratio of the signal power (sum of squares) over noise power
# Here, "singal" is defined as the audio part that passes a specific band filter, whilel "noise" is defined as the residual (noise = audio - signal).
def get_snr(audio, sampling_rate, lowest_freq, highest_freq):
	B, A = scipy.signal.butter(N=6, Wn=(lowest_freq,highest_freq), fs=sampling_rate, btype='bandpass', analog=False)
	signal_part = scipy.signal.lfilter(B, A, audio)
	noise_part = audio-signal_part
	snr = numpy.nanmean(signal_part**2)/numpy.nanmean(noise_part**2)
	return snr


for w,filepath in enumerate(wav_filepaths):
	print("Loading wav '%s'.."%(filepath))
	audio, sampling_rate = librosa.load(filepath, sr=RESAMPLING_RATE)
	if(audio.ndim>1): audio = audio[:,0] # keep only first channel, if there are multiple
	snr = get_snr(audio, sampling_rate, lowest_freq=100, highest_freq=1000)
	print("  --> SNR = %g"%(snr))