Audio Classifier Tutorial

Refer to this Google Colab Notebook

Prerequisite Knowledge

Sampling Frequency:# of samples per second eg. if the sampling frequency is 44100 hertz, a recording with a duration of 60 seconds will contain 2,646,000samples.

Dataset (UrbanSound8K)

• 10 Folders in total

• 9 Folders to train

• 1 Folder to test

Formatting the Data

Create wrapper class called UrbanSoundDataset that is a subclass of torch.utils.data.Dataset:

Fields (stores ____ of audio files when an object is instantiated)

• File names

• Labels

• Folder number

Methods: __getitem__

• Used to load and format data

• Uses torchaudio.load()

  • Converts .wav files to tensors

  • Returns tuple (tensor, sampling frequency of audio file - 44100Hz)

• Uses torchaudio.transforms.DownmixMono()

  • Dataset has audio in 2 channels, this makes it into 1 channel

• Network takes input size of 32,000 samples (only 0.7 seconds), but most audio files in dataset have 100,000 samples)

  • Downsample to 8000Hz (32,000 samples now spans 4 seconds)

    • Achieved by taking every 5th sample of original audio tensor

      • Not every audio tensor long enough to handle downsampling (these tensors will be padded with 0s)

    • Minimum length that won't require padding is 160,000 samples (32,000*5)