Lecture 1: Getting started with python and audio¶

Created by Mark Hasegawa-Johnson, 2023 August 14. This notebook is in the public domain; please re-use any portion of it in any way that you wish.

This window is a Jupyter notebook. The code is python. To create signals, I'm using numpy: https://numpy.org. To use audio, I'm following the directions on https://musicinformationretrieval.com/ipython_audio.html.

1. Tone¶

First, let's create a pure tone, at the frequency of middle C on the piano.

In [1]:
import numpy as np
sr = 8000 # sampling rate, [samples/second]
t = np.arange(sr)/sr # labels for the time axis
middle_C = 255 # middle C on the piano is about 255 Hz
tone = np.sin(2*np.pi*middle_C*t)  # a tone at middle C

Let's view the waveform.

In [2]:
%matplotlib inline
import matplotlib.pyplot as plt
plt.figure(figsize=(14, 5))
plt.plot(t[:500],tone[:500])
plt.xlabel('Time (seconds)')
plt.title('255Hz Pure Tone')
Out[2]:
Text(0.5, 1.0, '255Hz Pure Tone')

Let's view the spectrogram, using librosa.

In [3]:
import librosa, librosa.display
TONE = librosa.stft(tone)
TONEdb = librosa.amplitude_to_db(abs(TONE))
plt.figure(figsize=(14, 5))
librosa.display.specshow(TONEdb, sr=sr, x_axis='time', y_axis='hz')
Out[3]:
<matplotlib.collections.QuadMesh at 0x7ff1dd4b1a50>

Now let's play the tone, using Ipython.display.audio.

In [4]:
import IPython.display as ipd
ipd.Audio(tone, rate=sr)
Out[4]:
Your browser does not support the audio element.

Now I want to write the waveform to a file, so I can open it using Praat or any other interface. For this I use soundfile: https://pysoundfile.readthedocs.io/en/latest/

In [5]:
import soundfile as sf
sf.write('tone_255.wav', tone, sr)

2. Click¶

Next, let me create a click. A click is mostly zeros, with just one high-amplitude sample in the middle.

In [6]:
click = np.zeros(8000)
click[4000] = 1000 # arbitrary high-amplitude sample
plt.figure(figsize=(14, 5))
plt.plot(t,click)
plt.xlabel('Time (seconds)')
plt.title('Click at 0.5 seconds, with amplitude of 1000')
Out[6]:
Text(0.5, 1.0, 'Click at 0.5 seconds, with amplitude of 1000')
In [7]:
CLICK = librosa.stft(click, hop_length=80, win_length=240)  # 30ms window length, with a 10ms hop between windows
CLICKdb = librosa.amplitude_to_db(abs(CLICK))
plt.figure(figsize=(14, 5))
librosa.display.specshow(CLICKdb, sr=sr, x_axis='time', y_axis='hz')
Out[7]:
<matplotlib.collections.QuadMesh at 0x7ff1dda5dea0>
In [8]:
ipd.Audio(click, rate=sr)
Out[8]:
Your browser does not support the audio element.
In [9]:
sf.write('click.wav', click, sr)

3. Noise¶

Finally, the third main type of signal: noise. We'll use the function np.random.randn, which generates 8000 independent samples of a zero-mean, unit-variance Gaussian random variable.

In [10]:
noise = np.random.randn(8000)
plt.figure(figsize=(14, 5))
plt.plot(t[:500],noise[:500])
plt.xlabel('Time (seconds)')
plt.title('Gaussian Noise, Mean=0.0, Variance=1.0')
Out[10]:
Text(0.5, 1.0, 'Gaussian Noise, Mean=0.0, Variance=1.0')
In [11]:
plt.figure(figsize=(14, 5))
NOISE = librosa.stft(noise, hop_length=80, win_length=240)
librosa.display.specshow(librosa.amplitude_to_db(abs(NOISE)), sr=sr, x_axis='time', y_axis='hz')
Out[11]:
<matplotlib.collections.QuadMesh at 0x7ff1ddd98220>
In [12]:
ipd.Audio(noise, rate=sr)
Out[12]:
Your browser does not support the audio element.
In [13]:
sf.write('noise.wav', noise, sr)

4. Next¶

Now go look at the same signals using Praat: https://uvafon.hum.uva.nl/praat/

In [ ]: