[AI from Scratch] Episode 304: Visualizing Waveform Data — How to Graph Audio Signals

Recap and Today’s Theme

Hello! In the previous episode, we explored LibROSA, a powerful Python library for audio processing. We learned how to load, play, and extract features from audio files easily. Now, it’s time to move on to a more visual approach by focusing on the visualization of waveform data.

Waveform data shows the changes in amplitude over time and is an essential tool for analyzing the structure and dynamics of audio signals. In this episode, we will learn how to visualize audio waveforms using LibROSA and Matplotlib to gain insights into the properties of sound.

What is Waveform Data?

Waveform data represents the variations in amplitude of an audio signal over time. Audio is generated through pressure changes in the air, and these changes are digitally captured as waveform data. By visualizing the waveform, we can intuitively understand the dynamics of sound, such as intensity changes and specific events like drum hits or speech pauses.

Visualizing Audio Waveforms with LibROSA

Let’s explore the steps to visualize waveform data using LibROSA and Matplotlib.

1. Loading Audio Files

First, we load the audio file using LibROSA. The audio data is returned as a waveform array (y) and the sampling rate (sr).

import librosa

# Path to the audio file
audio_path = 'example.wav'

# Load the audio file
y, sr = librosa.load(audio_path, sr=None)

• librosa.load(audio_path, sr=None): This loads the audio file and returns the waveform data (y) and the original sampling rate (sr). Setting sr=None ensures the audio is loaded at its original sampling rate.

2. Visualizing the Waveform

LibROSA provides the librosa.display module, which makes it easy to visualize audio data as a waveform.

import librosa.display
import matplotlib.pyplot as plt

# Visualize the waveform
plt.figure(figsize=(10, 4))
librosa.display.waveshow(y, sr=sr)
plt.title('Waveform')
plt.xlabel('Time (seconds)')
plt.ylabel('Amplitude')
plt.show()

• librosa.display.waveshow(): This function displays the waveform data along a time axis. By specifying y and sr, the waveform is correctly scaled.
• Labels: We use plt.xlabel() and plt.ylabel() to label the x-axis as “Time (seconds)” and the y-axis as “Amplitude,” making the graph more understandable.

Executing this code will produce a graph that shows the audio’s amplitude changes over time, allowing you to see the structure of the sound.

Detailed Analysis with Waveform Visualization

By visualizing the waveform, you can pinpoint specific moments of interest in the audio, such as sudden increases in loudness or silent periods. This is useful for editing or analyzing audio where you need to focus on particular sections.

Comparing Multiple Waveforms

If you want to compare different audio files or the effects of processing on the same file, you can visualize multiple waveforms on the same graph. Here’s an example where we compare the original audio with a time-stretched version:

# Time-stretch the audio (change speed)
y_fast = librosa.effects.time_stretch(y, 1.5)

# Compare multiple waveforms
plt.figure(figsize=(10, 6))

# Original waveform
plt.subplot(2, 1, 1)
librosa.display.waveshow(y, sr=sr)
plt.title('Original Waveform')
plt.xlabel('Time (seconds)')
plt.ylabel('Amplitude')

# Time-stretched waveform
plt.subplot(2, 1, 2)
librosa.display.waveshow(y_fast, sr=sr)
plt.title('Time-Stretched Waveform (1.5x speed)')
plt.xlabel('Time (seconds)')
plt.ylabel('Amplitude')

plt.tight_layout()
plt.show()

• librosa.effects.time_stretch(): This function stretches or compresses the audio in time. In this example, we increase the speed by 1.5 times.
• plt.subplot(): This allows us to display multiple graphs in a single figure for easy comparison.

This comparison helps visually identify how audio manipulations, like time-stretching, affect the waveform.

Zooming in on Specific Sections of Audio

For longer audio files, you may want to zoom in and visualize only a specific section. Here’s how to extract and display a portion of the waveform:

# Display a specific range (e.g., first 5 seconds)
start_sample = 0
end_sample = sr * 5  # Sampling rate * number of seconds

plt.figure(figsize=(10, 4))
librosa.display.waveshow(y[start_sample:end_sample], sr=sr)
plt.title('Waveform (0-5 seconds)')
plt.xlabel('Time (seconds)')
plt.ylabel('Amplitude')
plt.show()

This code extracts the first 5 seconds of the audio and visualizes just that portion. This is useful when you need to focus on a specific event, such as a musical note or a sudden sound.

Spectral Analysis Using Short-Time Fourier Transform (STFT)

While waveforms show the changes in amplitude over time, they do not reveal much about the frequency content of the audio. To analyze the frequency components, we can apply a Short-Time Fourier Transform (STFT) and display a spectrogram.

# Short-Time Fourier Transform (STFT)
D = librosa.stft(y)

# Convert the amplitude spectrum to decibel scale
S_db = librosa.amplitude_to_db(abs(D))

# Display the spectrogram
plt.figure(figsize=(10, 4))
librosa.display.specshow(S_db, sr=sr, x_axis='time', y_axis='log')
plt.colorbar(format='%+2.0f dB')
plt.title('Spectrogram')
plt.xlabel('Time (seconds)')
plt.ylabel('Frequency (Hz)')
plt.show()

The spectrogram provides a visual representation of how the frequency content of the audio changes over time, allowing you to identify characteristics like pitch, tone, and noise.

Practical Example: Comparing Voice and Instrumental Sounds

Waveform analysis is useful when comparing different types of sounds. For example, you can visualize and compare the waveforms of human speech and guitar sounds:

# Load human voice and guitar sound files
voice_path = 'voice.wav'
guitar_path = 'guitar.wav'

y_voice, sr_voice = librosa.load(voice_path, sr=None)
y_guitar, sr_guitar = librosa.load(guitar_path, sr=None)

# Compare waveforms
plt.figure(figsize=(10, 8))

# Voice waveform
plt.subplot(2, 1, 1)
librosa.display.waveshow(y_voice, sr=sr_voice)
plt.title('Voice Waveform')
plt.xlabel('Time (seconds)')
plt.ylabel('Amplitude')

# Guitar waveform
plt.subplot(2, 1, 2)
librosa.display.waveshow(y_guitar, sr=sr_guitar)
plt.title('Guitar Waveform')
plt.xlabel('Time (seconds)')
plt.ylabel('Amplitude')

plt.tight_layout()
plt.show()

This visual comparison helps distinguish the unique characteristics of different audio sources, such as the smoother patterns of a voice compared to the more repetitive patterns of guitar strumming.

Summary

In this episode, we focused on visualizing waveform data, a fundamental technique for analyzing audio signals. By using LibROSA and Matplotlib, we can easily visualize audio waveforms, compare different audio files, and zoom in on specific sections. These skills are essential for understanding the structure of sound and will help in further audio processing tasks, such as frequency analysis.

Next Episode Preview

In the next episode, we will explore spectrograms in detail, learning how to analyze the frequency components of audio over time. This will allow us to delve deeper into the audio characteristics that are not visible in waveform data.

---

Notes

• Waveform Data: A graphical representation of audio signal amplitude over time, used to analyze sound structure and events.
• Short-Time Fourier Transform (STFT): A technique for converting audio signals into the frequency domain to analyze their spectral content【659†source】.