Skip to main content

Audio Files with Librosa

Code and audio for this article can be found here:
https://git.arts.ac.uk/ipavlov/WikiMisc/blob/main/AudioProcessing.ipynb

Loading audio files

Librosa is a Python libray created for working with audio data. It's both easy to understand and has an extensive feuture list.
By default Librosa supports all the popular audio file extensions, like WAV, OGG, MP3, and FLAC.

import librosa
import librosa.display

filename = librosa.ex('trumpet') #Loads sample audio file
y, sr = librosa.load(filename)

y — is a NumPy matrix that contains audio time series. If audio file is mono it will be one-dimensional vector, if audio file is stereo it will be two-dimensional, and so on.
sr — is audio file's sampling rate.

Playing audio in Jupyter notebook

Using the code bellow you will be able to play audio inside of your notebook. The player is very basic, but will be enough for simpler projects.

from IPython.display import Audio

Audio(data=y, rate=sr)

Vizualising audio files

To vizualise our audio files we can use Matplotlib.

import matplotlib.pyplot as plt

plt.figure(figsize=(12, 4))
librosa.display.waveshow(y, sr=sr) #This functions will dipslay audio file's waveform.
plt.show()

You can use code bellow to vizualize audio file's spectrogram.

fig, ax = plt.subplots()

S = librosa.feature.melspectrogram(y=y, sr=sr, n_mels=128,
                                    fmax=8000)

S_dB = librosa.power_to_db(S, ref=np.max)
img = librosa.display.specshow(S_dB, x_axis='time',
                         y_axis='mel', sr=sr,
                         fmax=8000, ax=ax)
fig.colorbar(img, ax=ax, format='%+2.0f dB')
ax.set(title='Mel-frequency spectrogram')

Working with multiple audio files

You can work multiple audio files. Just assign the time series for each file to a different variable or create a list of audio time series.
The code bellow looks at all .wav files in audio folder.

from glob import glob

audio_filepaths = []

for filename in glob('./audio/*.wav'): 
    audio_filepaths += [filename]
    
audio_filepaths

For AI&ML applications often times our audio data needs to be of uniform length wise.length. To do this we can paddpad them.

padded_audio_files = []

max_allowed_length = 32000

for audio_filepath in audio_filepaths:

    y_voice, sr_voice = librosa.load(audio_filepath)
    
    if len(y_voice) > max_allowed_length:
        raise ValueError("data length cannot exceed padding length.")
    elif len(y_voice) < max_allowed_length:
        embedded_data = np.zeros(max_allowed_length)
        offset = np.random.randint(low = 0, high = max_allowed_length - len(y_voice))
        embedded_data[offset:offset+len(y_voice)] = y_voice
    elif len(y_voice) == max_allowed_length:
        embedded_data = y_voice
        pass
    
    padded_audio_files += [embedded_data]
    
padded_audio_files = np.array(padded_audio_files)

padded_audio_files.shape

You can also append multiple audio files together. Let's do it with our padded files.

long_number = padded_audio_files.flatten()
long_number.shape

The waveform of a new audio will look like this:

Note: Keep track of audio file shapes, espcially along the 2nd axis.

Advanced Audio Analysis

Librosa has a wide range of tools for audio analyses. Some examples are included bellow.
Short-time Fourier transform:

S = np.abs(librosa.stft(y))

S.shape

Decibel analysis:

D = librosa.amplitude_to_db(librosa.stft(y), ref=np.max)

D.shape

Spectral flux:

S = np.abs(librosa.stft(y))
times = librosa.times_like(S)
fig, ax = plt.subplots(nrows=2, sharex=True)
librosa.display.specshow(librosa.amplitude_to_db(S, ref=np.max),
                         y_axis='log', x_axis='time', ax=ax[0])
ax[0].set(title='Power spectrogram')
ax[0].label_outer()


onset_env = librosa.onset.onset_strength(y=y, sr=sr)
ax[1].plot(times, 2 + onset_env / onset_env.max(), alpha=0.8,
           label='Mean (mel)')

onset_env = librosa.onset.onset_strength(y=y, sr=sr,
                                         aggregate=np.median,
                                         fmax=8000, n_mels=256)
ax[1].plot(times, 1 + onset_env / onset_env.max(), alpha=0.8,
           label='Median (custom mel)')

C = np.abs(librosa.cqt(y=y, sr=sr))
onset_env = librosa.onset.onset_strength(sr=sr, S=librosa.amplitude_to_db(C, ref=np.max))
ax[1].plot(times, onset_env / onset_env.max(), alpha=0.8,
         label='Mean (CQT)')
ax[1].legend()
ax[1].set(ylabel='Normalized strength', yticks=[])

Filters

As with all Python libraries, to unlock the full potential of librosa they need to be used with other libraries. The code bellow shows you how to apply a butter filter to audio signal, with a help of SciPy.

scipy.signal.butter

fig, ax = plt.subplots(nrows=2, figsize=(12, 4), constrained_layout=True)

ax[0].set(title='Normal waveform')

librosa.display.waveshow(y, sr=sr, ax=ax[0])

sos = signal.butter(17, 150, 'hp', fs=1000, output='sos')
filtered = signal.sosfilt(sos, y)

ax[1].set(title='Filtered waveform')

librosa.display.waveshow(filtered, sr=sr, ax=ax[1])
Back to top