Generating spectrograms with arecord, sox and ffmpeg/avconv

Create a spectrogram (Image)

The following command records a WAV file using ALSA (Device 2.0). Sox is piped into the process here — which is optional — but can be used to apply filters such as noise reduction or normalization to the resulting file:

arecord -D hw:2,0 -r 32000 -f S16_LE -c 1 -t wav | sox -t wav -c 1 -L -b 16 -r 32000 - test.wav

Once you have the WAV file, generate a spectrogram image with sox:

sox test.wav -n spectrogram -o image.png

A spectrogram image can also be created directly from a video file using ffmpeg:

ffmpeg -i test.avi -lavfi showspectrumpic=s=hd480:legend=0,format=yuv420p out.png

For older systems using avconv instead of ffmpeg:

avconv -i test.avi -lavfi showspectrumpic=s=hd480:legend=0,format=yuv420p out.png

Create a spectrogram (Video)

To generate a scrolling spectrogram video from a video file, use ffmpeg's showspectrum filter:

ffmpeg -i test.avi -lavfi showspectrum=s=1280x480:slide=1,format=yuv420p -c:v libx264 out.avi

This renders the audio spectrum as a moving video, saved as out.avi.

Alternatives with mplayer and sox

You can also generate a spectrogram from a video using mplayer to extract the audio first:

mplayer test.mp4 -ao pcm:file=/dev/stdout -vo null > test.wav

Then generate the spectrogram from the extracted WAV file:

sox test.wav -n spectrogram -o test.png

Comparing Spectrograms

Spectrograms created this way can be directly compared. However, even with seemingly identical WAV files, this comparison can be tricky.

To make differences more visible, creating a spectrogram diff is recommended.

First, create a diff WAV file by mixing the two sources with inverted polarity:

sox -m -v 1 source1.wav -v -1 source2.wav diff.wav

Then generate the spectrogram of the difference:

sox diff.wav -n spectrogram -o diff.png

If the two files are identical, the result will be silence and the spectrogram will be nearly empty. Any audible differences show up as frequency content in the diff.