# -*- coding: utf-8 -*-

"""Untitled0.ipynb

Automatically generated by Colaboratory.

Original file is located at

https://colab.research.google.com/drive/1LZ9BX53fhzO6o-zbbmju_Mi6HV1yu7GK

"""

import librosa

import matplotlib.pyplot as plt

import librosa.display

#part1 - preprocessing the signal

#---------------------------------------------------------------------------

#A audio(Music-Jazz) is played for 30 secs and recorded.

#15 lakh plus samples are collected and sampling frequency is 48000.

#Therefore in one second 48000 samples are collected that means

#the total length of signal in time domain is approximately 32

#secs. for analysation we need to have approximately only 50,000

#samples therefore we need to downsample the signal by

#1516541/50,0000 == 30.33 or aproximately 31.

x, sr = librosa.load('/content/jazz-mp3.mp3')

print(sr)

print(len(x))

#length comes out to be 696663, far greater than 50,000. We see that Sr is 22050 meaning that in 1 second

#22050 samples are collected, this shows that the music ran for 699663/22050 secs == 31.73 secs which is true.

#to reduce the no of samples approximately we have to change the sampling rate.

#To find the new sampling rate

audio_time = len(x)//sr

new_sr = 50000//audio_time

print(new_sr)

#new_sr is equal to approx 1582.544

#Load the signal again but this with the new sr

x, sr = librosa.load('/content/jazz-mp3.mp3', sr=new_sr)

print(len(x))

#no of samples now is 50000 exact.

#plotting the graph

plt.figure(figsize=(14, 5))

librosa.display.waveplot(x, sr=new_sr)

#display Spectrogram

#to display the spectogram-change of frequency with time.

X = librosa.stft(x)

Xdb = librosa.amplitude_to_db(abs(X))

plt.figure(figsize=(14, 5))

librosa.display.specshow(Xdb, sr=sr, x_axis='time', y_axis='hz')

plt.colorbar()

#----------------------------------------------------------------------

#part2 - extracting features

#--------------------------------------------------------------------------

#1 finding Zero crossing point

zero_crossings = librosa.zero_crossings(x[0:50000], pad=False)

print(sum(zero_crossings))

#To visualise the zero crossing rate we zoom in the time v/s amplitude graph

#Zooming in

n0 = 3000

n1 = 3005

plt.figure(figsize=(14, 5))

plt.plot(x[n0:n1])

plt.grid()

#after zooming in by taking the window between 0 to 5 secs we see that the graph crosses 0.000 three times

#thus Zero_crossing_rate is 3 thus the result.

zero_crossings = librosa.zero_crossings(x[0:5], pad=False)

print(sum(zero_crossings))

#2. Spectral centroid

import sklearn

spectral_centroids = librosa.feature.spectral_centroid(x, sr=sr)[0]

#this calculates spectral centroids for each frame in the spectrum.

spectral_centroids.shape

frames = range(len(spectral_centroids))

t = librosa.frames_to_time(frames)

def normalize(x, axis=0):

return sklearn.preprocessing.minmax_scale(x, axis=axis)

#librosa.display.waveplot(x, sr=sr, alpha=0.4)

#plt.plot(t, normalize(spectral_centroids), color='r')

#By observing the first graph we observe that spectral centroids are prominent at the beginning.To combat this we add a constant at the beginning

#and plot again.

spectral_centroids = librosa.feature.spectral_centroid(x+0.5, sr=sr)[0]

#librosa.display.waveplot(x, sr=sr, alpha=0.4)#alpha denotes the opacity of the grapn which is less than in order to visualise.

#plt.plot(t, normalize(spectral_centroids), color='b')

#3 Spectral Bandwidth

spectral_bandwidth_2 = librosa.feature.spectral_bandwidth(x+0.01, sr=sr)[0]

librosa.display.waveplot(x, sr=sr, alpha=0.4)

plt.plot(t, normalize(spectral_bandwidth_2), color='g')

#3. Spectral RollOff

spectral_rolloff = librosa.feature.spectral_rolloff(x+0.01, sr=sr)[0]

librosa.display.waveplot(x, sr=sr, alpha=0.4)

plt.plot(t, normalize(spectral_rolloff), color='y')

#4. Tempo

tempo = librosa.beat.tempo(x, sr=sr)

print(tempo.item(0))

#this gives the tempo of the audio signal which is approximately 188.90 for jazz music.

#-------------------------------------------------------------------------------------------------