A Novel Method for Music Retrieval using Chord Progression

Transcription

1 Australia Joural of Basic ad Applied Scieces, 10(2) Special 2016, Pages: AUSTRALIAN JOURNAL OF BASIC AND APPLIED SCIENCES ISSN: EISSN: Joural home page: A Novel Method for Music Retrieval usig Chord Progressio 1 R.L. Jasmie, 2 G. David RAJ, 3 G.V. Uma 1 Research Scholar, IST Departmet, Collegee of Egieerig, Aa Uiversity, Cheai. 2 Research Scholar, IST Departmet, Collegee of Egieerig, Aa Uiversity, Cheai. 3 Professor, IST Departmet, College of Egieerig, Aa Uiversity, Cheai. Address For Correspodece: R.L. Jasmie, 1 Research Scholar, IST Departmet, College of Egieerig, Aa Uiversity, Cheai. A R T I C L E I N F O Article history: Received 04 December 2015 Accepted 22 Jauary 2016 Available olie 14 February 2016 Keywords: Chords, music classificatio, Scale A B S T R A C T Music iformatio retrieval (MIR) systems allow users to search collectios of music. Surfig ca be performed, for example, by meas of metadata usig keywords to describe music or the actual such that the melodic cote of a musical piece. Music specificatio with respect to its scale is a emergig area i music iformatio retrieval. A key problem i music specificatio is how to efficietly extract audio features for high level classificatio. May types of features have bee used for music classificatio icludig low level features such as timber ad temporal features ad mid-level features such as chord, rhythm, ad beat for music aalysis. Most preset work i music iformatio retrieval survey allows music via low-level features like Mel frequecy cepstral coefficiet (MFCC) ad other spectral coefficiets. Although, low-level features are isufficiet for may applicatios cosiderig to the sigal characteristics comparatively to the sematic cotet of music. O the cotrary, mid-level features such as Chord sequece, which describes harmoic progressio ad toal structure of music, is oe of the most importat mid-level features of music. With chord sequece, sogs that are similar i various aspects ca be idetified ad retrieved more effectively. This paper focus o chord recogitio ad its applicatios. INTRODUCTION There are large quatity of text, audio, video ad other documets available o the iteret, o about ay subject. Search is oe of the most importat user iterface elemets i ay large website. There are two ways of searchig the iformatio: (1) To use a search egies. (2) To browse directories orgaized by categories. Iformatio retrieval (IR) is the process of gatherig applicable iformatio to a query. IR is differet from data retrieval, which is about to fid small amout of data i a databases with a give structure. I IR systems, the iformatio is ot structured; it is cotaied i free form i text or i multimedia cotet. There are may types of iformatio retrieval they are: (1) Speech retrieval, which deals speech cotiuously trascribed maually by automated speech recogitio (ASR). (2) Cross laguage revival uses a query i a laguage say Eglish ad perceive documet i other laguages like Chiese ad Russia. (3) Questio aswerig IR system, which retrieve aswers from a body of text. (4) Image retrieval, which fids image o a theme. (5) Music retrieval, which fids a piece whe a user hums a melody or eter the otes of a musical theme. Music is a widely ejoyed cotet type, existig i may multifaceted represetatios. With the digital iformatio ere, a lot of digital music iformatio has apparetly available at the user s figertips. However the abudace of iformatio is too wide-ragig ad too various to compose, ispect ad preset i a cosistet ad huma maer, motivatig the developmet of Music Iformatio Retrieval (Music-IRcollectio ad media replay o trasferable devices, fruitful retrieval techiques. Due to the fast growth of digital music ad Ope Access Joural Published BY AENSI Publicatio 2016 AENSI Publisher All rights reserved This work is licesed uder the Creative Commos Attributio Iteratioal Licese (CC BY). /4.0/ To Cite This Article: R.L. Jasmie, G. David RAJ, G.V. Uma, A Novel Method for Music Retrieval usig Chord Progressio Aust. J. Basic & Appl. Sci., 10(2): , 2016

2 278 R.L. Jasmie et al., 2016 Australia Joural of Basic ad Applied Scieces, 10(2) Special 2016, Pages: maagemet of music is eeded i the digital era. Most curret work i music iformatio retrieval peruse music via low-level features however, low-level features are isufficiet for umerous requisitio sice they are equivalet to the sigal characteristics preferably tha the sematic cotet of music. Separatio of audio Chord Recogitio Scale Learig Retrieval of similar music Comp are Musical Kowled ge Musical Database O the perverse, mid-level characteristics for example as chord, rhythm, ad istrumetatio cotai rich iformatio for music aalysis. Based o the iformatio of the music some music orieted system is available such as last.fm, sheetmusicarchive.et, musicsheetplus.com, stagepass.com ad pado. However such a approach suffers from drawbacks: (1) First a ewly added music must be recogized. (2) Secod upopular music recordigs may be recogized. Chord Research Backgroud: A chord is a set of toes played simultaeously. A chord progressio is a sequece of chords over time ad is what describes the harmoy of a piece (Mitchell, T.M., 2010). Automatic chord recogitio is the process of extractig a chord progressio from a audio file. These chord sequeces are used by musicias as lead sheets summaries cotaiig chords, melody, ad lyrics as well as by researchers for tasks such as key detectio, gere classificatio ad lyric iterpretatio. Performig chord aalysis by had is time cosumig, proe to huma error, ad requires two or more traied experts. sequece, which arrates characteristics series ad toal compositio of music, is oe of the most importat mid-level features of music. Sice music characteristics series is strogly related to distiguish sesatio, alike chord successio ca be oticed i sogs that are adjacet i gere, emotio, etc. With chord sequece, sogs that are similar i various aspects ca be distiguished ad retrieved more effectively. I this paper, a chord recogitio method is proposed. This is what makes automatic chord recogitio a importat area of research (Schacht, A ad W. Sommer, 2009). The two mai steps of automatic chord recogitio are feature extractio ad patter matchig. Feature extractio is the process of extractig useful iformatio from audio files, ad patter matchig is how chord labels are applied to that data. There are may challeges ecoutered by systems that process audio sigals. There are backgroud oises, percussio istrumets ad other uwated toes i audio recordigs. It is also difficult to distiguish whe chords chage ad to lie these poits up exactly with the beat. Preprocessig helps elimiate uwated iformatio from the audio files before or durig the feature extractio step, depedig o the system. Existig perusal o chord idetificatio is predomiatly based o HMM (Posser, J., 2008). The geeral procedure is described as follows. First, the iput audio is cleaved ito successive frames ad trasformed to frequecy domai. The, feature vectors for chord idetificatio are retrieved. Sice chord are composed of harmoy formed by multiple otes or pitches, the twelve dimesioal pitch class profile 9PcP) feature vector or chroma vector, which represets the itesity of twelve semitoe pitch class, is mostly used (Hu, X ad J.S. Dowie, 2010l; Upadhye, R ad V.H. Sahasrabuddhe, 1992; Schacht, A ad W. Sommer, 2009). Very little work has take place i the area of applyig techiques from computatioal musicology. The special work doe by Sahasrabudha et al. (Ghias, A et al, 1995) gives the ability to idetify raga i a melodic type of compositio. The system was based o assigig a raga to is earest seve ote scale. The seve ote scale, as is also the case i Wester music, is cosidered a typical full scale. Proposed Architecture: The harmoic progressio ad toal structure of music ca be described by a chord sequece. Harmoic progressio ca iterrupt chord sequece from which gere, emotio ca be observed. With chord sequece, similar sogs i various scales ca be idetified ad retrieved more successfully.

3 279 R.L. Jasmie et al., 2016 Australia Joural of Basic ad Applied Scieces, 10(2) Special 2016, Pages: Separatio Of Audio From The Music: Imagie you are stadig i a busy street amog a crowd of people. You ca glea traffic soud, footfall of people adjacet, the bleepig of a rambler crossig, your mobile phoe rigig, ad co-workers at the back of you havig a coferece. Eve though all these differet soud sources, you have a pretty good idea of what is goig o aroud you. It is more tha just a mess of overlappig oise, ad if you try hard you ca cocetrate o oe of these sources if it is importat to you such as the coversatio behid you. There are two mai obstacles for their comprehesive acquisitio determied o the sceario. The mai limitatio i some cases is: (1) High latecy ad computatioal cost. (2) I other cases the quality of the results is isufficiet. Audio editig software is software which allows to maage ad perceipitate of audio data. Audio editig software ca be implemeted completely or partly as library, as computer approach, as Web approach or as a loadable kerel module. There are differet types of audio editig software they are: (1) Adobe Auditio (2) AVS Audio Editor (3) Diamod Cut (4) Wave Lab (5) Music Maker Dowload ad istall AVS Audio Editor: Ope A Audio File: To ope a audio file that you ited to chop pop the Ope butto o the Home tab of the Ribbo Commad Bar. Fig. 2: To ope a audio file Select the Necessary Segmets: To select the parts of your audio file that you eed to have as separate files apply markers. Start by idicatig the first segmet: pop the required positio withi the Waveform. Display with the left mouse butto ad compress the Add Marker butto o the Edit tab. The start of the file will be the begiig of the opeig chuch ad the added marker idicates its ed. Fig. 3: Settig markers for the selected music Split Your Audio ito Several Files:

4 280 R.L. Jasmie et al., 2016 Australia Joural of Basic ad Applied Scieces, 10(2) Special 2016, Pages: As soo as all the required markers are set click the Split File by Markers lik o the Edit tab. the audio will be split ito various files accordig to the markers set. They will be called as the iput file ame + the sequece umber ad put the Files tab of the result ad Filters pael. Save the Resultig Tracks: To retai all the derived tracks utilize the Save All optio of the File tab. If you do ot eed all the tracks double click the essetial file at the result ad Filters pael ad use the Save as optio to save this follow with ay desigatio, place, format whichh ca suit your eeds best. Fig. 4: Savig the audio track. Represetig the Chords Of Sogs: To represet the chords of a sog is described briefly. A good represetatio will make the search efficiet ad allow adaptable user iterfaces es ad tools to be costructed. A sog is a combiatio of differet keys formed to make a tue. They are costructed by the pitch values of the otes. The occurrece of each ote i the sog is the duratio of all otes icludig the chords. Fig. 5: represetatio of chords i a sog Similarity Relatios of Chords: The otes o the keyboard are give by comparig them to the otes of the alphabets. The keyboard otes ca be amed as c-d-e-f-g-a-b. Each ote differs with each other i soud. The seve otes of keyboard repeat themselves agai ad agai. That the otes soud the same but the pitch differs. For example if c ote is played ad moved to the right util it fids the ext c, both otes souds the same but oe is higher tha the other. A stadard semiprofessioal music keyboard has forty eight keys. Oe of these twelve set of otes is techically called a octave. Wester music is based o frequecies of the keys. A octave is dividig ito twelve equal itervals such that the frequecy ratio of two eighborig itervals is the same. This iterval is called a semi toe. There are twelve mutually exclusive half otes i the system. I wester music middle c octave that is also called the middle c scale etc. starts from the first white key set to 240 Hz. Oce the octave is figured out, it is easy to idetify the first key of this octave9set to 240 Hz. The below table give the frequecies of each scale. Recogizig Chords: The basic of wester music is the recogitio of chords. If a tue stays i oe key, the the basic chords are formed by the combiatio of differet otes o the scale. Much sog uses the three chords based o the basics (i), the quarter (IV) ad the fifth (V) ). Sometimes a fourth ote is added which is a third above the highest ote of the triad. This gives a seveth chord. Whe a mior third is used it ca result i the use of a ote which is ot actually part of the idegieousl scale. For sample, the C cosiderable triad is CEG ad a small third above the G is Bb which is ot part of the C major scale.

5 281 R.L. Jasmie et al., 2016 Australia Joural of Basic ad Applied Scieces, 10(2) Special 2016, Pages: Table 1: List of frequecies for otes A HZ A# \ Bb HZ B HZ C HZ C# \ Db HZ D HZ D# \ Eb HZ E HZ F HZ F# \ Gb HZ G HZ G#\ Ab HZ A HZ The problem of fidig the key piece of chord is called key fidig. The method behid the key fidig is that the frequecies of each key are compared. Whe a sog matches a particular key the frequecies distace betwee the chords ad the toe of the key are relatively small. The formula is give by: S=αr(I)+r(IV)+r(V)+(β if the irst chord matched the key β if the irst chord matched the key Here r deotes the scale of the key. A deotes the frequecy of the other key i the scale; B supports the first ad last key of the chord. Cmi gmi=cmi f7=c7 f#mi7=c #mi7 f7=cf fmi=cmi f=c f7=c7 cmi dmi=cmi emi=cmi c a7=c7 Fig. 6: Tree structure of chord sequece Coclusio: A chord i a Idia music is a very complex structure. The sequece of otes used to play the sogs is based o the chords. The proposed architecture aims to build the database so that it icludes a large umber of examples i each chord. The proposed system aalyzes the sequece of scales for the chord idetified. Here, it ot oly idetifies the chords, but it also tags them accordig to their scales. The scope of the model ca also be exteded to mood based audio retrieval. The combiatio of audio ad lyrical cotet is a crucial factor for future mood classificatio system. REFERENCES Bradley, M.M ad P.J. Lag, Affective orms for Eglish words (ANEW): stimuli, istructio maual ad affective ratigs, The Ceter for Research i Psychophysiology, Uiversity of Florida. Garcia-Silva A. et al, 2012.Review of the state the art: Discoverig ad associatig sematics to tags i folksoomies, The Kowledge Egieerig Review, 27(1): Ghias, A et al, 1995.Query by Hummig Musical Iformatio Retrieval i a Audio Database, I Proceedigs of ACM Multimedia, pp: Golder, S.A ad B.A. Huberma, 2006, Usage patters of collaborative taggig systems, Joural of Iformatio ad Sciece, 32(2): Hu, X ad J.S. Dowie, Whe lyrics outperform audio for music mood classificatio: a feature aalysis, I the Proceedigs of ISMIR, pp: 1-6. Hu, X et al., 2009, Lyric text miig i music mood classificatio, I Proceedigs of the 10th ISMIR Coferece, pp: Hu, X., creatig a simplified music mood classificatio groud truth set, I Proceedigs of ISMIR, pp: Jusli, P.N ad Sloboda J.Am, Hadbook of Music ad Emotio: Theory, Research, Applicatios,

6 282 R.L. Jasmie et al., 2016 Australia Joural of Basic ad Applied Scieces, 10(2) Special 2016, Pages: Oxford, MS, USA: Oxford Uiversity Press. Mitchell, T.M., 2010, predictig huma brai activity associated with the meaigs of ous, Huma Brai Mappig, 32(30): Peterse, M.K et al, Latet sematics as cogitive compoet, I Proceedigs of 2d Iteratioal Workshop o Cogitive Iformatio Processig CIP. Posser, J., The europhysiological bases of emotio: a fmri study of the affective circumflex usig emotio-deotig words, Huma Brai Mappig, 30(3): Schacht, A ad W. Sommer, Emotios i word ad face processig early ad late cortical resposes, Brai ad Cogitio, 20(2): Sog, Y et al. 2011, Automatic tag recommedatio algorithms for social recommeder systems, ACM Trasactios o the Web, 5(1): 4. Upadhye, R ad V.H. Sahasrabuddhe, O the Computatioal Model of Raga Music o Idia, I Proceedigs of 10th Europea Coferece o Artificial Itelligece, Viea.