Skip to contents

Managing audio in R

Source: vignettes/manage_audio.Rmd
manage_audio.Rmd

Managing audio in R with SonicScrewdriveR

Reading audio files

Several functions are available to read audio files into R, including the readWave() and readMP3() functions from the tuneR package, as well as tools from the package av. SonicScrewdriveR simplifies reading audio files by providing a single wrapper for these functions, readAudio(), which can read audio files in a variety of formats, including WAV, MP3, and FLAC.

filename <- system.file("extdata", "AUDIOMOTH.WAV", package="sonicscrewdriver")
w <- readAudio(filename)

Performing analyses on channels

Some existing functions only operate on a single channel at a time. This may cause unnecessarily complex workflows when bulk analysing files with different numbers of channels. The allChannels() function applies a function to each channel and returns a list of analyses. This technique allows for the same analysis to be performed on each channel, without reference to the number of channels in the audio file. Optionally, a cluster can be specified to process channels on separate processor cores to increase analysis speed.

Windowing

It is often desirable to process a long audio file in chunks. The windowing() function can be used to split an audio file into overlapping or non-overlapping windows. This function may be particularly useful for processing long Wave files in a memory-efficient manner. Optionally, a cluster can be specified to process windows on separate processor cores to increase processing speed.

In order to demonstrate windowing() we first define a simple function that draws a rectangle around the windowed region of a sound file.

drawWindow <- function( wave, start, window.length) {
  rect(start, -1, start+window.length, 1, col= rgb(0,0,1.0,alpha=0.5))
}

We can then show the windows generated if the window.length is 44100 samples, and the window.overlap is 0.

# Create a 5 second sine wave of 1Hz
w <- tuneR::sine(1, duration=5*44100)

plot(w@left, type="l", xlab="Time (samples)", ylab="Amplitude")

windowing(w, window.length=44100, window.overlap = 0, FUN=drawWindow)

The entire audio file is analysed in chunks of 44100 samples, with no overlap between windows. The drawWindow() function is applied to each window, and the result is plotted on top of the oscillogram of the original audio file.

The window.overlap parameter can be adjusted so that the windows overlap by a certain number of samples.

plot(w@left, type="l", xlab="Time (samples)", ylab="Amplitude")

windowing(w, window.length=44100, window.overlap = 44100/2, FUN=drawWindow)

Alternatively, a negative value of window.overlap can be used to take regularly-spaced samples from the audio file.

plot(w@left, type="l", xlab="Time (samples)", ylab="Amplitude")

windowing(w, window.length=44100, window.overlap = -44100, FUN=drawWindow)

The bind.wave parameter can be used to combine the results of the windowing function into a single Wave object (if FUN returns a Wave object).

In the example below we use windowing() to add noise to sections of a sine wave.

w <- tuneR::sine(1, duration=5*44100)

addNoise <- function(w, start, window.length) {
  nw <- tuneR::noise("white", duration=length(w@left), samp.rate=w@samp.rate, pcm=w@pcm, bit=w@bit)
  rw <- w + nw/max(nw@left) # Scale noise to the amplitude of the sine wave
  return(rw)
}

o <- windowing(w, window.length=44100, window.overlap = -44100, FUN=addNoise, bind.wave=TRUE)
plot(o@left, type="l", xlab="Time (samples)", ylab="Amplitude")

PseudoWave objects

TaggedWave workflow

The techniques above can be applied to the generic Wave and WaveMC objects from the tuneR package.

The TaggedWave class extends the Wave class from the tuneR package so that it can include extended metadata and the results of analyses. This allows for the storage of additional information about the audio file, such as the location and time of recording, and the results of analyses. The tagWave() function can be used to tag a Wave or WaveMC object with additional metadata.

In addition, combined with new classes WaveAugment, WaveFilter, and WaveAnalyse it is possible to create a self-documenting pipeline for audio processing and analysis (that is also compatible with the R pipe).