NOTE: I haven't written this, merely found it on the internet and ported to python 3.
- https://shivama205.medium.com/audio-signals-comparison-23e431ed2207 "Audio signals: Comparison"
- https://gist.github.com/shivama205/5578f999a9c88112f5d042ebb83e54d5 scripts from the article
Related projects:
- https://acoustid.org/chromaprint fpcalc
- numpy
Usage:
Sample files captured from a streaming source without exact start, duration but are the same song:
$ ./compare.py -i file1.mp3 -o file2.mp3
Calculating fingerprint by fpcalc for file1.mp3
Calculating fingerprint by fpcalc for file2.mp3
File A: file1.mp3
File B: file2.mp3
Match with correlation of 63.74% at offset 55
$ ./compare.py -i file2.mp3 -o file1.mp3
Calculating fingerprint by fpcalc for file2.mp3
Calculating fingerprint by fpcalc for file1.mp3
File A: file2.mp3
File B: file2.mp3
Match with correlation of 63.74% at offset -5
For some files the swapped order may not lead to the same results due to offset
or the way the fpcalc fingerprint is generated (see help).
$ ./compare.py -i file2.mp3 -o file3.mp3
Calculating fingerprint by fpcalc for file2.mp3
Calculating fingerprint by fpcalc for file3.mp3
File A: file2.mp3
File B: file3.mp3
Match with correlation of 93.01% at offset -24
$ ./compare.py
Calculating fingerprint by fpcalc for file1.mp3
Calculating fingerprint by fpcalc for file3.mp3
File A: file1.mp3
File B: file3.mp3
Match with correlation of 63.96% at offset 31
Internally the fingerprint is generated by fpcalc -length 500, cached
versions can be produced by fpcalc-gen.
Changes:
- port to python3
- print the similary as percents
- print input files on separate lines
- support precalculated fingerprint in
file.mp3.fpcalc