Plagiarized Audio Identification Using Audio

Fingerprinting
Er. M Ahmed Siddiqui Arshi
M.H.Saboo Siddik College of Engineering M.H.Saboo Siddik College of Engineering
[email protected] [email protected]

Ansari Hera Shaikh Swaleha

Abstract
As a result of recent technological advancement we Fingerprints are short summaries of multimedia
currently have a close to endless provider of musical system content. The aim of fingerprinting is to
content. With their additional and a lot of demand it's produce quick and reliable suggests that of
become quite vital to research them. The potency to protection, management and classification of
seek out the precise song from an oversized audio multimedia system content. It's the same as Human
information is especially a really vital task for fingerprint is an unique identity of individual.
variety of applications. During this paper we are Similarly an audio fingerprint is employed for
aiming to provides a detail view regarding our topic. recognizing an audio clip.[1]
This method can take away plagiarism of audio from
an audio database using the technique of audio In general, the process perform must have the
fingerprinting. It'll eventually save users time and subsequent properties:
memory. It'll also avoid redundancy of audio files in
a very large database.  Robustness: the fingerprints ensuing from
degraded versions of an audio ought to end
in an equivalent or a minimum of similar
1. Introduction fingerprints with reference to the previous
fingerprint .
Recent technological achievements have resulted in a
very larger access to digital music. Digital libraries  Pair-wise independence: if 2 audios are
contain innumerable songs whereas personal mp3 perceptually totally different, the
players permit us to relish our music wherever we fingerprints from 2 audios ought to be
are. Digital radio, websites and recommendation significantly totally different.
software system are even serving to us find the artists
that charm to us. Thanks to this endless provider of  Database search efficiency: for the sensible
music, there's a growing interest in new ways of applications with a large-scale fingerprint
filtering the content of audio information. Among the information, quick sound unit search is
field of Music information Retrieval there's a desire crucial.[2][3]
to develop Tools and applications which will give
economical identification and looking. The potency The audio fingerprinting technique has several
to seek out the precise song from an oversized audio applications in broadcast observation, automatic
information is especially a really vital task for variety generation of play list of radio, TV, or web
of applications. broadcasts etc. Another application can be stated as
looking out (retrieving) the name of the creator and
title of the song given a brief clip of an audio,
filtering technology for file sharing and so on. Since
an audio contains several features, extracting a

1
unique audio fingerprint is the most significant task 1. Within the 1st phase all those songs are searched
that describes every song. and retrieved that might probably contain the
candidate’s fingerprint.
2. Within the second part every matched songs full
2. Literature Survey fingerprint is fetched from the database and is
compared with the query fingerprint.
Haitsma and Kalker introduced an classification If the song matches with the query fingerprint then
technique that works on the principle of search table. it's retrieved and also the songs hash count is
A search table consist of all the combinations of currently raised by one unit i.e. If the songs hash
audio fingerprints. The strategy lets the entries within count is zero it's currently 1 after the search and every
the search table point to the song or songs wherever time it's accrued by one unit when it's retrieved when
the respective sub-fingerprint value occurs. A sub- looking. After change the hash count of the searched
fingerprint worth will occur at multiple points in a song, its position is updated within the database
song or multiple songs, the song pointers are kept in a according to its hash count, the a lot of the hash count
linked list. So one sub-fingerprint will purpose to the higher are its position i.e. The highest most
multiple pointers at an equivalent time and so all the position, therefore whenever this songs query comes
256 blocks of fingerprint block are compared with the next time then it'll be retrieved from the updated
entries in database.[4] position and not from the recent position, which can
decrease the time of retrieval thereby increasing the
This downside is removed to some extent within the speed of looking out.[6]
next work. 585 Avery Wang introduced the theme of
landmark hashing. The fundamental operation of this In easy terms, if you would like to match audio files
theme is that every audio track is analyzed to seek out by their sensory activity equality, you must produce
outstanding onsets focused in frequency, since these the so referred to as "fingerprints" (similar to human
onsets are possibly to be preserved in noise and fingerprints, that unambiguously describe a human
distortion. These onsets are shaped into pairs, identity), and see if sets of those objects, gathered
parameterized by the frequencies of the peaks and from totally different audio sources match or not.
also the time in between them. These values are Logically, similar audio objects ought to generate
quantized to provide a comparatively sizable amount similar fingerprints, whereas totally different files
of distinct landmark hashes (about one million in ought to emanate in contrast to signatures. One of the
number). Parameters are tuned to provide around 20- wants for these fingerprints is that they must act as
50 landmarks per second. Each reference track is "hashes", so as to deal with format variations, noise,
represented by the (many hundreds) landmarks it "loudness", etc. The simplified idea of audio process
contains, and also the times at that they occur is kept are often visualized below.
in an inverted index. Equally to spot a query it's
regenerated to landmarks. This technique once more
suffers from same downside i.e. It will be seen that it
contains regarding one million landmark hashes
however again the matter is that to look a audio file
it'll head to an equivalent position wherever the audio
file is kept and so every time when a song is
requested it'll undergo an equivalent method. It can
be seen that the matter of accuracy is resolved to an
honest extent however still a meg landmarks is
extremely high and so extends the search domain still
to an oversized space. To beat the above 2 issues we
utilize the thought of Bunching with the help of Hash
Count.[5]

3. Proposed System
Figure 1: Working of the system
The projected search rule contains 2 phases as
represented below: Basically, audio process algorithms give the
flexibility to link short, untagged snippets of audio
content to corresponding data this content (it's all

2
regarding finding distinctive characteristics of a song A special mention here to Prof. Z. A. Usmani
which will be later used to recall an unknown sample, (H.O.D., Computer Engineering Department,
or notice a replica within the information of already MHSSCOE) for his valuable support. We are also
processed songs). These kinds of algorithms are often thankful to all staff members of Computer
utilized in a range of various applications: Department, without whom the completion of this
 Automatic population of the metadata of a report would have been impossible.
song
 Identifying the presently enjoying song This entire journey would not have been possible
 Managing effect libraries without the efforts put in by our guides, Er. Mohd
 Monitoring radio broadcasts Ahmed. They have been a constant source of
encouragement and guidance through the entire
With all the benefits that the sound process system semester.
has, there are many challenges that it's to deal with.
one among them could be a vast information to look This acknowledgment would indeed be incomplete
(imagine YouTube's information of video content without rendering our sincere gratitude to our family.
that's monitored for audio copyright issues). They have always been a pillar of strength and
Normally, every song can generate an enormous support in our past and current endeavors.
quantity of fingerprints (in the represented rule, the
granularity of each of them goes to be 1.48 sec,
therefore we'll have about two hundred objects for a
song). lastly, there'll be a requirement for using an 6. References
economical search algorithm that works well, once [1]. Duplicate Song Detection using Audio Fingerprinting
the solution is scale.[7][8] for Consumer Electronics Devices – IEEE Paper
Published in:
4. Conclusion Consumer Electronics, 2006. ISCE '06. 2006 IEEE
Tenth International Symposium on
If you have lots of music on your computer, you may [2]. Ke, Y., D. Hoiem, and R. Sukthankar. 2005. Computer
have several versions of the same song. Maybe the vision for music identification. IEEE Computer
music is identical, but the filenames are different. Or Society Conference on Computer Vision and Pattern
maybe the song is the same, but you have multiple Recognition 597–604.
copies each recorded at a different bit rate. This
application allows you to do is to perform a search [3]. Arslan, L.M., Hansen, J.L., 1999. Selective training
for duplicates based only on song title, for instance. for hidden markov models with applications to speech
This frees up disk space on your computer, your classi®cation.IEEE Trans. Speech and Audio
Processing 7 (1), 46±54.
music device and can eliminate hearing the same
song more than once. The aim of this project is to [4]. Chen, W.-Y., Chen, S.-H., Lin, C.-J., 1996. A speech
introduce to the concept of audio recognition and recognition method based on the sequential multi-
analysis. The proof of concept will successfully layer perceptrons. Neural Networks 9 (4), 655±669
detect the duplicate songs from huge collection of
audio files with the help of plagiarized audio [5]. Meinard Müller, Frank Kurth, and Michael
identification using audio fingerprinting. Clausen.Audio matching via chromabased statistical
features. In Proceedings of the International
Conference on Music Information Retrieval (ISMIR),
5. Acknowledgement pages 288–295, London, UK, 2005.
No project can be completed without the support of a
[6]. Meinard Müller and Frank Kurth.Towards structural
lot of people. We are concluding our project work by
analysis of audio recordings in the presence of musical
submitting this report, we reflect upon all the times variations.EURASIP Journal on Advances in Signal
when we needed support in various forms and were Processing, 2007(1), 2007.
lucky enough to receive it.
[7]. Jouni Paulus, Meinard Müller, and
We wish to express our sincere gratitude to our AnssiKlapuri.Audio-based music structure analysis. In
principal Dr.Mohiuddin Ahmed, M. H. Saboo Proceedings of the International Society for Music
Siddik College of Engineering, Mumbai, for Information Retrieval Conference (ISMIR), pages
providing the facilities to carry out the project work. 625–636, Utrecht, The Netherlands, 2010.

3
[8]. RakeshAgrawal, King-Ip Lin, Harpreet S. Sawhney,
and Kyuseok Shim. Fast similarity search in the
presence of noise, scaling, and translation in time-
series databases. In Proceedings of the International
Conference on Very Large Data Bases (VLDB), pages
490–501, Zurich, Switzerland, 1995.