A small command-line tool to find similar audio files

First, install the Chromaprint fingerprinting library by Lukáš Lalinský. (The library itself depends on an FFT library, but it's smart enough to use an algorithm from software you probably already have installed; see the Chromaprint page for details)

Then you can install this library:

pip install audiomatch

To do things fast audiomatch requires C compiler and Python headers to be installed. You can skip compilation by setting AUDIOMATCH_NO_EXTENSIONS environment variable:

AUDIOMATCH_NO_EXTENSIONS=1 pip install audiomatch

You can avoid installing all this libraries on your computer and run everything in docker:

docker run --rm -v "$(pwd)":/tmp fdooch/audiomatch "/tmp/*"

Suppose, we have a directory with Nirvana songs:

$ ls demo
All Apologies (In Utero).m4a           Dumb (Unplugged in NYC).m4a
All Apologies (Unplugged in NYC).m4a   Pennyroyal Tea (In Utero).m4a
Dumb (In Utero).m4a                    Pennyroyal Tea (Solo Acoustic).mp3
Dumb (Radio Appearance, 1991).mp3      Pennyroyal Tea (Unplugged in NYC).m4a

Let's find out which files sound similar:

$ audiomatch --length 300 ./demo
These files sound similar:

./demo/All Apologies (In Utero).m4a
./demo/All Apologies (Unplugged in NYC).m4a

---

./demo/Dumb (In Utero).m4a
./demo/Dumb (Unplugged in NYC).m4a

---

./demo/Pennyroyal Tea (In Utero).m4a
./demo/Pennyroyal Tea (Solo Acoustic).mp3
./demo/Pennyroyal Tea (Unplugged in NYC).m4a

Note #1: input audio files should be at least 10 seconds long

Note #2: in some rare cases false positives are possible

What's happening here is that audiomatch takes all audio files from the directory and compares them with each other.

You can also compare file with another file, file and directory, or directory to directory. If you need to, you can provide glob-style patterns, but don't forget to quote it, because otherwise shell expanded it for you. For example, let's compare all .mp3 files with .m4a files:

$ audiomatch  "./demo/*.mp3" "./demo/*.m4a"
These files sound similar:

../demo/Pennyroyal Tea (Solo Acoustic).mp3
../demo/Pennyroyal Tea (Unplugged in NYC).m4a

This time, audiomatch took all files with .mp3 extension and compare them with all files with .m4a extension.

Note, how there is no In Utero version in the output. The reason it is present in the previous output, because it actually similar with Unplugged version and then transitive law applies: if a = b and b = c, then a = c.

The --length specifies how many seconds to take for analysis from the song. Default value is 120 and it is good enough to find exactly the same song, but maybe in different quality. However, for a more complicated cases like same song played in different tempo the more input we have the more accurate results are.

By default, audiomatch looks for files with .m4a, mp3, .caf extensions. In theory, audio formats supported by ffmpeg also supported by audiomatch. You can tell to audiomatch to look for a specific format by using --extension flag:

$ audiomatch -e .ogg -e .wav ./demo
Not enough input files.

Indeed, we tried to compare files with .ogg and .wav extension, but there are no such files in the demo directory.

I play guitar and do recordings from time to time mainly with Voice Memos on iPhone. Over the years, I have hundreds of recordings like that and I though it would be cool to find all the similar ones and see how I progress over the years.

That's why I wrote this library.

• Chromaprint and pyacoustid libraries
• Example: How to compare fingerprints
• Example: How to compare shifted fingerprints (note: the code is a little bit weird)
• Explanation: How to compare fingerprints
• Popcount in Python with benchmarks