Qfp
Qfp is a python library for creating audio fingerprints that are robust to alterations in pitch and speed. This method is ideal for ID'ing music from recordings such as DJ sets where the tracks are commonly played at a different pitch or speed than the original recording. Qfp is an implementation of the audio fingerprinting/recognition algorithms detailed in a 2016 academic paper by Reinhard Sonnleitner and Gerhard Widmer [1].
NOTE: This is a fork of https://github.com/mbortnyck/qfp/ which incorporates:
- multiple bug fixes
- Python 3 compatibilty
- modern packaging with Poetry
This is all in service of having a convient audio fingerprinting service
for a website I'm working on.
Quickstart
Install by downloading it from PyPI.
pip install qfp
You can create a fingerprint from your reference audio...
from qfp import ReferenceFingerprint
fp_r = ReferenceFingerprint("Prince_-_Kiss.mp3")
fp_r.create()
...or a query fingerprint from an audio clip that you wish to identify.
from qfp import QueryFingerprint
fp_q = QueryFingerprint("unknown_audio.wav")
fp_q.create()
The QfpDB can store reference fingerprints...
from qfp.db import QfpDB
db = QfpDB()
db.store(fp_r, "Prince - Kiss")
... and look up query fingerprints.
fp_q = QueryFingerprint("kiss_pitched_up.mp3")
fp_q.create()
db.query(fp_q)
print(fp_q.matches)
[Match(record=u'Prince - Kiss', offset=0, vScore=0.7077922077922078)]
Qfp currently accepts recordings in any format that FFmpeg can handle.
Dependencies
FFmpeg - https://github.com/FFmpeg/FFmpeg
NumPy - https://github.com/numpy/numpy
Pydub - https://github.com/jiaaro/pydub
SciPy - https://github.com/scipy/scipy
SQLite - https://www.sqlite.org/
[1] R. Sonnleitner and G. Widmer, "Robust quad-based audio fingerprinting," IEEE/ACM Transactions on Audio, Speech and Language Processing (TASLP), vol. 24, no. 3, pp. 409–421, Jan. 2016. [Online]. Available: http://ieeexplore.ieee.org/abstract/document/7358094/. Accessed: Nov. 15, 2016.