/qfp

Quad-based audio fingerprinting and recognition in Python

Primary LanguagePython

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].

Quickstart

Install by cloning the repository.

git clone https://github.com/mbortnyck/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.