Android app for automatic music composition

ANDROID APP FOR AUTOMATIC MUSIC COMPOSITION Ma ina S. Nie o Ana M. Ba bancho Isabel Ba bancho Lo enzo J. Ta d´ on ATIC Resea ch G oup, Dp o. Ingenie ´ ıa de Comunicaciones, E.T.S.I. Telecomunicaci´ on, Uni e sidad de M´ alaga, Andaluc´ ıa Tech, Campus Uni e si a io de Tea inos s/n, 29071, M´ alaga, Spain [email p o ec ed], [email p o ec ed], [email p o ec ed], [email p o ec ed] ABSTRACT In his con ibu ion, we p esen an And oid App ha gene - a es music au oma ically eplica ing a speci ic con empo- a y gen e, speci ically ock music. The composi ion sys- em is based on he analysis o a se o songs o he desi ed gen e. The analysis esul s he au oma ic composi ion al- go i hm equi es a e expo ed o a JSON ile ha can be eadily impo ed and pa sed. The And oid App de eloped c ea es a symbolic ep esen a ion o he music gene a ed o d ums, bass and hy hm gui a ha can be expo ed o a MIDI ile o audio wa e o m. The e alua ion esul s show ha he au oma ic composi ions gene a ed a e no dis in- guished amongs o he exce p s c ea ed by human com- pose s. 1. INTRODUCTION Since he beginning o he compu e a ea, he e has been a g owing in e es on he c ea ion o a by machines. Many musicians and compose s ha e used ma hema ical me h- ods and se s o ules o c ea e musical pieces, in ac , al- go i hmic composi ion could be de ined as he employ- men o a se o ules ha combines musical elemen s o o m a music piece [1], [2], [3]. In his con ibu ion, we use a p obabilis ic model o he cha ac e iza ion o music lea ned om music samples [2]. A e he musical model is lea ned, a music composi o ha ollows he ules in [2] is implemen ed o And oid de ices. Ou composi o gen- e a es new ock music exce p s wi h h ee di e en ins u- men s: d ums, bass and hy hmic gui a . 2. SYSTEM DESCRIPTION The s uc u e o he And oid App o au oma ic music composi ion de eloped is shown in Fig. 1. In his igu e, i can be seen ha he de eloped sys em cosis s o wo di - e en pa s: analysis and pa e n c ea ion s age and com- posi ion sys em. The i s one was implemen ed in Ma lab and he second one o And oid de ices in o de o c ea e a mobile applica ion. c Ma ina S. Nie o, Ana M. Ba bancho, Isabel Ba bancho, Lo enzo J. Ta d´ on. Licensed unde a C ea i e Commons A ibu ion 4.0 In e na ional License (CC BY 4.0). A ibu ion: Ma ina S. Nie o, Ana M. Ba bancho, Isabel Ba bancho, Lo enzo J. Ta d´ on. “And oid App o au oma ic music composi ion”, 19 h In e na ional Socie y o Music In- o ma ion Re ie al Con e ence, Pa is, F ance, 2018. T aining Da ase Analysis and Pa e n C ea ion Composi ion sys em Pa ame e s selec ed by he use Ou pu Song JSON ile Figu e 1. Au oma ic music composi ion scheme. 2.1 Analysis and pa e n c ea ion Among he di e en possible musical gen es, ock was se- lec ed. The inpu o he analysis and pa e n segmen a ion subsys em is a da ase wi h 25 songs belonging o he clas- sical ock music gen e. Each song in he da ase , is cha ac- e ized by h ee MIDI iles, one o each o he ins umen s conside ed in he au oma ic music compose (d ums, bass and hy hm gui a ). The analysis and pa e n disco e y is ca ied ou by implemen ing a speci ic ool in Ma lab which is a modi ica ion o he one desc ibed in [2], o in- clude cho d pa e ns. Cho d pa e ns a e needed in o de o compose music o d ums and hy hm gui a . Fig. 2 shows a schema ic o he analysis and pa e n c ea ion subsys em. T aining Da ase Res s inclusion Ba segmen a ion JSON ile Pa e nDa abase Rhy hmic pa e ns Pi ch pa e ns Cho d pa e ns Figu e 2. Analysis and pa e n c ea ion subsys em. Th ee di e en ypes o pa e ns a e lea ned, wi h each indi idual pa e n co esponding o a ba : hy hmic pa - e ns: du a ion o he no es in bea s; pi ch pa e ns: pi ch con ou in o ma ion; and cho d pa e ns: no es sounding simul aneously o each o he hy hmic igu es. Cho d pa e ns cons i u e a new addi ion wi h espec o he ap- p oach in [2]. I mus be said ha d ums a e conside ed no uned ins umen s and he cho d pa e ns o d ums sound a each bea . 2.2 Composi ion sys em The composi ion sys em i s pe o ms he selec ion o he hy hmic s uc u es. The hy hmic s uc u es las ou ba s and hey a e andomly selec ed be ween among he ollow- ing: AAAA, ABAB, ABBA, AABB, AABC, ABCA and ABCD. The leng h o inal composi ions a e a combina- ion o hese ou -ba s uc u es. Once he gene al hy h- mic s uc u es a e de ined, he ack o each ins umen is gene a ed. Each ins umen is ea ed in a sligh ly di e - en way due o he di e ences be ween d ums, bass and hy hmic gui a . In all ins umen , i s hy hmic, pi ch and cho d pa e ns a e selec ed. The selec ions a e made in a p obabilis ic way [2] wi h he ollowing pa icula i ies o each ins umen : •D um ack gene a ion: since d ums a e no uned ins umen s, cho ds a e simply mapped in o a d um ki . •Bass ack gene a ion: he pi ch is co ec ed in o de o i he melody o he cho d p og ession gene a ed. •Rhy hmic gui a ack gene a ion: pi ch and cho d a e co ec ed in o de o i gui a and bass. 3. ANDROID APP INTERFACE The main sc een o he de eloped applica ion is p esen ed in Fig. 4. This sc een shows he use op ions: he use can selec he ins umen (s) o include in he composi ion: d ums, bass, gui a ; key: majo o mino ; empo in bpm, om 110 o 150 bpm); and composi ion leng h: 4 ba s o comple e song wi h he s uc u e shown in 3. Once he composi ion pa ame e s a e selec ed, he use mus p ess he bu on w i e song. A e he song is com- posed, i can be played in he phone o expo ed ia e-mail (MIDI ile) o Wha sApp (*.wa ile). In oduc ion (4 ba s) Ve se (8 ba s) Cho us (8 ba s) Ve se (8 ba s) Cho us (8 ba s) Figu e 3. Song s uc u e. 4. SYSTEM EVALUATION A su ey has been designed o e alua e he quali y o he au oma ic compose . The su ey includes h ee ques ions, o each ques ion, h ee di e en music exce p s a e con- side ed: wo o hem a e ex ac ed om popula ock songs and he emaining one was composed by he applica ion desc ibed. The pa icipan s a e eques ed o iden i y he exce p s ha in hei opinion was c ea ed by an au oma ic composi ion algo i hm. A ou h possible choice is al- lowed: ”I don’ know”. The su ey also inqui es whe he he subjec s ecognized any o he songs and, in such case, iden i y hem in o de o e alua e i he pa icipan s musi- cal knowledge a ec ed hei answe s. A o al o 97 people answe ed he su ey; he esul s a e shown in Table 1. As i can be seen om he esul s, in Figu e 4. And oid App o music composi ion. Iden i ica ion No Iden i ica ion Ques ion 1 47.42 % 52.58 % Ques ion 2 37.11 % 62.89 % Ques ion 3 17.53 % 82.47 % To al 34.02 %65.98 % Table 1. Su ey esul s. he i s ques ion almos hal o he pa icipan s iden i ied he au oma ic compose ; when we e iewed he answe s ega ding he ecogni ion o he exce p s, se e al pa ici- pan s indica ed ha hey ecognized one o he samples in- cluded in ha ques ion, which is consis en wi h he esul s ob ained. 5. CONCLUSIONS An au oma ic music composi ion App has been p esen ed. This App gene a es ock music and includes d um, bass and hy hmic gui a . In he nex mon hs i will be published in he Google Play S o e. A subjec i e e alua ion shows ha he exce p s c ea ed a e no dis inguished amongs o h- e s c ea ed by human compose s. 6. ACKNOWLEDGEMENTS This wo k has been unded by he Minis e io de Economa y Compe i i idad o he Spanish Go e nmen unde P ojec No. TIN2016-75866-C3-2-R. The wo k has been done a Uni e sidad de Mlaga, Campus de Excelencia In e na- cional Andaluca Tech. 7. REFERENCES [1] D. Cope. Algo i hmic composi ion [ e]de ined. In Panel Discussion in ICMC P oceedings, pages 23–25, 1993. [2] C. Roig, L.J. Ta don, I. Ba bancho, and A.M. Ba - bancho. Au oma ic melody composi ion based on a p obabilis ic model o music s yle and ha monic ules. Knowledge-Based Sys ems, (71):419–434, 2014. [3] C. Roig, L.J. Ta don, I. Ba bancho, and A.M. Ba ban- cho. A non-homogeneous bea -based ha mony ma ko model. Knowledge-Based Sys ems, (142):85–94, 2018.