To recognize long recorded files.

[shuvuu]

Hello, I am new to this project and trying to learn and solve my problem. My problem is to recognize 1 hour long live FM recording files. My database belongs ~ 2500+ mp3 files, limit: 30 seconds! randomly selected 20 files & combined them & saved first 1 hour into test file. Tried to recognize what songs where in this file.

`

!/usr/bin/python

import sys import warnings import json, datetime import audioread from pydub import AudioSegment encoding="utf-8" reload(sys) sys.setdefaultencoding('utf8') warnings.filterwarnings("ignore")

from dejavu import Dejavu from dejavu.recognize import FileRecognizer

with open("dejavu.cnf.SAMPLE") as f: config = json.load(f)

if name == 'main':

fingerprint_limit : 30

# database: ~ 2500 mp3
# fingerprints: ~ 37+ million

djv = Dejavu(config)
recognizer = FileRecognizer(djv)

# randomly selected & combined ~20 songs from ~2500 mp3 files 
# cut the beginning 1 hour
filename = "1.mp3" 
input_song = AudioSegment.from_mp3(filename)
duration = len(input_song)/1000
start = 0
chunk_length = 20

while (start < duration):
    #20-30 seconds long chunk file. pydub
    chunk = input_song[start * 1000:start * 1000 + chunk_length*1000]
    chunk.export("tmp.mp3", format="mp3")

    res_song = recognizer.recognize_file("tmp.mp3")

    if (res_song):
        print "Result: %s" % repr(res_song).decode("unicode-escape") 
        # match_time ~ 10 - 180+ seconds !!! 
        # print correctly unicode song names

         ''' song_duration - this is the new column I added to songs table.             
        Related changes made to fingerprint_file, insert_song, ... functions.
        I got duration value using :

            try:
                with audioread.audio_open(filepath) as f:
                duration = f.duration
            except:
                duration = 0            
        '''
        start = start + res_song['song_duration'] - 5
    else:
        start = start + 25 # because my fingerprint_limit : 30

` -- this is the code I used. Hope you understand. Total process time ~ 15 - 30 minutes (defends on chunk_length)

the problem :

1. Slow : fingerprint.fingerprint(samples, Fs=Fs) - this function takes almost 1 minute to work on this ~20 seconds chunk files?!?
2. I think similar songs allowed. In my case first 30 seconds are similar. Then false results returned. Of course if we match whole song then correct results should return.

And I want to know how to fingerprint with offset or it's staring position? Now it starts from beginning, I want it start from different points. For example: I want to fingerprint mp3 file's - from 00:00:45 to 00:01:15 - part OR last 30 seconds. If it's possible then I could store unique songs' parts to database to avoid Problem 2.

[thesunlover]

there is already an issue for this topic ;)