Audio Datasets

torchaudio provides easy access to common, publicly accessible datasets. Please refer to the official documentation for the list of available datasets.

# When running this tutorial in Google Colab, install the required packages
# with the following.
# !pip install torchaudio

import torch
import torchaudio

print(torch.__version__)
print(torchaudio.__version__)

Out:

1.11.0+cpu
0.11.0+cpu

Preparing data and utility functions (skip this section)

# @title Prepare data and utility functions. {display-mode: "form"}
# @markdown
# @markdown You do not need to look into this cell.
# @markdown Just execute once and you are good to go.

# -------------------------------------------------------------------------------
# Preparation of data and helper functions.
# -------------------------------------------------------------------------------
import multiprocessing
import os

import matplotlib.pyplot as plt
from IPython.display import Audio, display


_SAMPLE_DIR = "_assets"
YESNO_DATASET_PATH = os.path.join(_SAMPLE_DIR, "yes_no")
os.makedirs(YESNO_DATASET_PATH, exist_ok=True)


def plot_specgram(waveform, sample_rate, title="Spectrogram", xlim=None):
    waveform = waveform.numpy()

    num_channels, num_frames = waveform.shape

    figure, axes = plt.subplots(num_channels, 1)
    if num_channels == 1:
        axes = [axes]
    for c in range(num_channels):
        axes[c].specgram(waveform[c], Fs=sample_rate)
        if num_channels > 1:
            axes[c].set_ylabel(f"Channel {c+1}")
        if xlim:
            axes[c].set_xlim(xlim)
    figure.suptitle(title)
    plt.show(block=False)


def play_audio(waveform, sample_rate):
    waveform = waveform.numpy()

    num_channels, num_frames = waveform.shape
    if num_channels == 1:
        display(Audio(waveform[0], rate=sample_rate))
    elif num_channels == 2:
        display(Audio((waveform[0], waveform[1]), rate=sample_rate))
    else:
        raise ValueError("Waveform with more than 2 channels are not supported.")

Here, we show how to use the torchaudio.datasets.YESNO() dataset.

dataset = torchaudio.datasets.YESNO(YESNO_DATASET_PATH, download=True)

for i in [1, 3, 5]:
    waveform, sample_rate, label = dataset[i]
    plot_specgram(waveform, sample_rate, title=f"Sample {i}: {label}")
    play_audio(waveform, sample_rate)

• 
• 
• 

Out:

  0%|          | 0.00/4.49M [00:00<?, ?B/s]
  1%|          | 32.0k/4.49M [00:00<00:25, 186kB/s]
  4%|4         | 200k/4.49M [00:00<00:07, 635kB/s]
 18%|#8        | 832k/4.49M [00:00<00:01, 1.97MB/s]
 64%|######4   | 2.88M/4.49M [00:00<00:00, 5.86MB/s]
100%|##########| 4.49M/4.49M [00:00<00:00, 5.74MB/s]
<IPython.lib.display.Audio object>
<IPython.lib.display.Audio object>
<IPython.lib.display.Audio object>