ketanhdoshi

import librosa

# Load the audio file
AUDIO_FILE = './audio.wav'
samples, sample_rate = librosa.load(AUDIO_FILE, sr=None)

ketanhdoshi

from scipy.io import wavfile
sample_rate, samples = wavfile.read(AUDIO_FILE)

ketanhdoshi

import librosa.display
import matplotlib.pyplot as plt

# x-axis has been converted to time using our sample rate. 
# matplotlib plt.plot(y), would output the same figure, but with sample 
# number on the x-axis instead of seconds
plt.figure(figsize=(14, 5))
librosa.display.waveplot(samples, sr=sample_rate)

ketanhdoshi

sgram = librosa.stft(samples)
librosa.display.specshow(sgram)

ketanhdoshi

# use the mel-scale instead of raw frequency
sgram_mag, _ = librosa.magphase(sgram)
mel_scale_sgram = librosa.feature.melspectrogram(S=sgram_mag, sr=sample_rate)
librosa.display.specshow(mel_scale_sgram)

ketanhdoshi

# use the decibel scale to get the final Mel Spectrogram
mel_sgram = librosa.amplitude_to_db(mel_scale_sgram, ref=np.min)
librosa.display.specshow(mel_sgram, sr=sample_rate, x_axis='time', y_axis='mel')
plt.colorbar(format='%+2.0f dB')

ketanhdoshi

print ('Example shape ', samples.shape, 'Sample rate ', sample_rate, 'Data type', type(samples))
print (samples[22400:22420])

ketanhdoshi

from IPython.display import Audio
Audio(AUDIO_FILE)

ketanhdoshi

#Spectrogram is a 2D numpy array
print(type(mel_sgram), mel_sgram.shape)
# <class 'numpy.ndarray'> (128, 134)

ketanhdoshi

import math, random
import torch
import torchaudio
from torchaudio import transforms
from IPython.display import Audio

class AudioUtil():
  # ----------------------------
  # Load an audio file. Return the signal as a tensor and the sample rate
  # ----------------------------