No full-text available
To read the full-text of this research,
you can request a copy directly from the authors.
Automatic Melodic Harmonization: An Overview, Challenges and Future Directions
Abstract
Automatic melodic harmonization tackles the assignment of harmony content (musical chords) over a given melody. Probabilistic approaches to melodic harmonization utilize statistical information derived from a training dataset, producing harmonies that encapsulate some harmonic characteristics of the training dataset. Training data is usually annotated symbolic musical notation. In addition to the obvious musicological interest, different machine learning approaches and algorithms have been proposed for such a task, strengthening thus the challenge of efficient & effective music information utilisation using probabilistic systems. Consequently, the aim of this chapter is to provide an overview of the specific research domain as well as to shed light on the subtasks that have arisen and since evolved. Finally, new trends and future directions are discussed along with the challenges which still remain unsolved.
... Automatic melody harmonization aims to build up a learning machine that can generate chord sequences to accompany a given melody [1,2]. In music, a chord is any harmonic set composed of three or more notes that are heard as if sounding simultaneously [3]. However, melody harmonization is a highly subjective task that is difficult for machines to learn to imitate humans or fit human preferences. ...
... Many approaches have been proposed for automatic melody harmonization [3], such as hidden Markov models (HMMs) [4,5,6] and genetic algorithm (GA)-based methods [7]. Recently, with the prevalence of deep learning models, some deep learning methods have emerged to deal with the melody harmonization problem [8]. ...
Coherence and interestingness are two criteria for evaluating the performance of melody harmonization, which aims to generate a chord progression from a symbolic melody. In this study, we apply the concept of orderless NADE, which takes the melody and its partially masked chord sequence as the input of the BiLSTM-based networks to learn the masked ground truth, to the training process. In addition, class weighting is used to compensate for some reasonable chord labels that are rarely seen in the training set. Consistent with the stochasticity in training, blocked Gibbs sampling with proper numbers of masking/generating loops is used in the inference phase to progressively trade the coherence of the generated chord sequence off against its interestingness. The experiments were conducted on a dataset of 18,005 melody/chord pairs. Our proposed model outperforms the state-of-the-art system MTHarmonizer in five of six different objective metrics based on chord/melody harmonicity and chord progression. The subjective test results with more than 100 participants also show the superiority of our model.
... In their overview paper, Makris et al. (2016) explain two types of melodic harmonization: chorale harmonization and accompaniment. Chorale harmonization creates four-part harmonies for classical music. ...
Systems such as Microsoft Songsmith automatically assign chords and harmony to a melody by minimizing the dissonance across all chord changes. Although this produces harmonious music, it is not what practicing musicians do. In this paper, I describe Harmonizer, a prototype system for melodic harmonization. Harmonizer uses Schoenberg's chart of regions as the underlying data structure that allows harmonization using several different methods. Because the chart reveals inter-chordal relationships, the harmonizations may be programmed to emphasize desired relationships. In the prototype Harmonizer, I also explore recent signal-processing methods that enable songwriters to easily input a melody by singing or by playing a musical instrument. The prototype Harmonizer is available on GitHub and a video demonstrating its distinctive harmonizations is on YouTube as explained in the Results section of the paper.
... International Journal of Computing accompaniment to all the rules of classical musical harmony [10]. We mean harmonization as the process of assignment musical chords to a given musical theme, presented as a melody. ...
The problem of musical accompaniment harmonization for adjacent transitional fragments in computer games is considered. A harmonization method based on a tonal model of the musical composition in conjunction with a genetic algorithm seeking the global extremum of the harmonic function is proposed. For the method under consideration, all operators of the genetic algorithm and fitness functions are defined. A scheme was developed for the formation of transitional musical compositions and a flowchart was proposed for using the genetic algorithm in its implementation. Using examples of well-known musical compositions, a study was conducted of the genetic algorithm features and its most appropriate parameters were found in the context of the problem under consideration. This allowed us to obtain expression for the computational costs of the proposed method, which are necessary for balancing the computational load in the computer between the game algorithm itself and the musical harmonization algorithm. Moreover, the study showed that using genetic algorithm allows even for a small number of iterations to achieve melody harmonization level no worse than professional composer achieved, and even exceed this level when performing extra iterations.
... A UTOMATIC melody harmonization, which finds a coherent chord sequence that fits the given notes in a melody, is an essential topic in music generation. This task, which imitates the harmonizing process, is important for understanding human composition [1]. It is also practical for commercial use since it can reduce barriers to creating music without expertise [2], [3]. ...
Recent deep learning approaches for melody harmonization have achieved remarkable performance by overcoming the uneven chord distributions of music data. However, most of these approaches
have not attempted to capture an original melodic structure and generate structured chord sequences with appropriate rhythms. Hence, we use a Transformer-based architecture that directly maps lower-level melody notes into a semantic higher-level chord sequence. In particular, we encode the binary piano roll of a melody into a note-based representation. Furthermore, we address the flexible generation of various chords with Transformer expanded with a VAE framework. We propose three Transformer-based melody harmonization models: 1) the standard Transformer-based model for the neural translation of a melody to chords (STHarm); 2) the variational Transformer-based model for learning the global representation of complete music (VTHarm); and 3) the regularized variational Transformer-based model for the controllable generation of chords (rVTHarm). Experimental results demonstrate that the proposed models generate more structured, diverse chord sequences than LSTM-based models.
... Melody harmonization is an application of the algorithmic composition [26], where the goal of some studies is to automatically generate a chord progression for a given melody [22,29,34], and others to compose a four-part chorale for a melody [14,18,21]. This paper focuses on the former. ...
Melody harmonization, namely generating a chord progression for a user-given melody, remains a challenging task to this day. Although previous neural network-based systems can effectively generate an appropriate chord progression for a melody, few studies focus on controllable melody harmonization, and none of them can generate flexible harmonic rhythms. To achieve harmonic rhythm-controllable melody harmonization, we propose AutoHarmonizer, a neural network-based melody harmonization system that can generate denser or sparser chord progressions with the use of a new sampling method for controllable generation proposed in this paper. This system mainly consists of two parts: a harmonic rhythm model provides coarse-grained chord onset information, while a chord model generates specific pitches for chords based on the given melody and the corresponding harmonic rhythm sequence previously generated. To evaluate the performance of AutoHarmonizer, we use nine metrics to compare the chord progressions from humans, the system proposed in this paper and the baseline. Experimental results show that AutoHarmonizer not only generates harmonic rhythms comparable to the human level, but generates chords with overall better quality than baseline at different settings. In addition, we use AutoHarmonizer to harmonize the Session Dataset (which were originally chordless), and ended with 40,925 traditional Irish folk songs with harmonies, named the Session Lead Sheet Dataset, which is the largest lead sheet dataset to date.
Melody harmonization, which involves generating a chord progression that complements a user-provided melody, continues to pose a significant challenge. A chord progression must not only be in harmony with the melody, but also interdependent on its rhythmic pattern. While previous neural network-based systems have been successful in producing chord progressions for given melodies, they have not adequately addressed controllable melody harmonization, nor have they focused on generating harmonic rhythms with flexibility in the rates or patterns of chord changes. This paper presents AutoHarmonizer, a novel system for harmonic density-controllable melody harmonization with such a flexible harmonic rhythm. AutoHarmonizer is equipped with an extensive vocabulary of 1462 chord types and can generate chord progressions that vary in harmonic density for a given melody. Experimental results indicate that the AutoHarmonizer-generated chord progressions exhibit a diverse range of harmonic rhythms and that the system’s controllable harmonic density is effective.
Coherence and interestingness are two criteria for evaluating the performance of melody harmonization, which aims to generate a chord progression from a symbolic melody. In this study, we apply the concept of orderless NADE, which takes the melody and its partially masked chord sequence as the input of the BiLSTM-based networks to learn the masked ground truth, to the training process. In addition, the class weights are used to compensate for some reasonable chord labels that are rarely seen in the training set. Consistent with the stochasticity in training, blocked Gibbs sampling with proper numbers of masking/generating loops is used in the inference phase to progressively trade the coherence of the generated chord sequence off against its interestingness. The experiments were conducted on a dataset of 18,005 melody/chord pairs. Our proposed model outperforms the state-of-the-art system MTHarmonizer in five of six different objective metrics based on chord/melody harmonicity and chord progression. The subjective test results with more than 100 participants also show the superiority of our model.
The task of automatic melody harmonization aims to build a model that generates a chord sequence as the harmonic accompaniment of a given multiple-bar melody sequence. In this paper, we present a comparative study evaluating the performance of canonical approaches to this task, including template matching, hidden Markov model, genetic algorithm and deep learning. The evaluation is conducted on a dataset of 9226 melody/chord pairs, considering 48 different triad chords. We report the result of an objective evaluation using six different metrics and a subjective study with 202 participants, showing that a deep learning method performs the best.
Melodic harmonisation deals with the assignment of harmony (chords) over a given melody. Probabilistic approaches to melodic harmonisation utilise statistical information derived from a training dataset to harmonise a melody. This paper proposes a probabilistic approach for the automatic generation of voice leading for the bass note on a set of given chords from different musical idioms; the chord sequences are assumed to be generated by another system. The proposed bass voice leading (BVL) probabilistic model is part of ongoing work, it is based on the hidden Markov model (HMM) and it determines the bass voice contour by observing the contour of the melodic line. The experimental results demonstrate that the proposed BVL method indeed efficiently captures (in a statistical sense) the characteristic BVL features of the examined musical idioms.
During the last decades, several methodologies have been proposed for the harmonization of a given melody with al-gorithmic means. Among the most successful are method-ologies that incorporate probabilistic mechanisms and sta-tistical learning, since they have the ability to generate har-monies that statistically adhere to the harmonic character-istics of the idiom that the training pieces belong to. The current paper discusses the utilization of a well–studied probabilistic methodology, the hidden Markov model (HMM), in combination with additional constraints that incorporate intermediate fixed–chord constraints. This work is moti-vated by the fact that some parts of a phrase (like the ca-dence) or a piece (e.g. points of modulation, peaks of ten-sion, intermediate cadences etc.) are characteristic about the phrase's or piece's idiomatic identity. The presented methodology allows to define and isolate such important parts/functions and include them as constraints in a proba-bilistic harmonization methodology. To this end, the con-strained HMM (CHMM) is developed, harnessed with the novel general chord type (GCT) representation, while the study focuses on examples that highlight the diversity that constraints introduce in harmonizations.
In this paper we focus on issues of harmonic representa-tion and computational analysis. A new idiom-independent representation is proposed of chord types that is appropriate for encoding tone simultaneities in any harmonic context (such as tonal, modal, jazz, octatonic, atonal). The General Chord Type (GCT) representation, allows the re-arrangement of the notes of a harmonic simultaneity such that abstract idiom-specific types of chords may be derived; this encoding is inspired by the standard roman numeral chord type labeling, but is more general and flexible. Given a consonance-dissonance classification of intervals (that reflects culturally-dependent notions of consonance/dissonance), and a scale, the GCT algorithm finds the maximal subset of notes of a given note simultaneity that contains only con-sonant intervals; this maximal subset forms the base upon which the chord type is built. The proposed representa-tion is ideal for hierarchic harmonic systems such as the tonal system and its many variations, but adjusts to any other harmonic system such as post-tonal, atonal music, or traditional polyphonic systems. The GCT representa-tion is applied to a small set of examples from diverse musical idioms, and its output is illustrated and analysed showing its potential, especially, for computational music analysis & music information retrieval.
We propose a system, the Continuator, that bridges the gap between two classes of traditionally incompatible musical systems: (1) interactive musical systems, limited in their ability to generate stylistically consistent material, and (2) music imitation systems, which are fundamentally not interactive. Our purpose is to allow musicians to extend their technical ability with stylistically consistent, automatically learnt material. This goal requires the ability for the system to build operational representations of musical styles in a real time context. Our approach is based on a Markov model of musical styles augmented to account for musical issues such as management of rhythm, beat, harmony, and imprecision. The resulting system is able to learn and generate music in any style, either in standalone mode, as continuations of musician’s input, or as interactive improvisation back up. Lastly, the very design of the system makes possible new modes of musical collaborative playing. We describe the architecture, implementation issues and experimentations conducted with the system in several real world contexts.
Most melody harmonization systems use the generative hidden Markov model (HMM), which model the relation between the hidden chords and the observed melody. Relations to other variables, such as the tonality or the metric structure, are handled by training multiple HMMs or are ignored. In this paper, we propose a discriminative means of combining multiple probabilistic models of various musical variables by means of model interpolation. We evaluate our models in terms of their cross-entropy and their performance in harmonization experiments. The proposed model offered higher chord root accuracy than the reference musicological rule-based harmonizer by up to 5% absolute.
The purpose of this paper is to present Robin, an algorithmic composer specifically designed for interactive situations. Users can interact in real time with the algorithmic composition by means of control strategies based on emotions. This study aims at providing a system for automatic music generation to be applied to interactive systems for music creation targeted at non-musicians. Robin adopts a rule-based approach to compose original tonal music in classical piano style. The first practical application of Robin is The Music Room, an interactive installation which enables people to compose tonal music in pairs by communicating emotion expressed by moving throughout a room.
For a system to be able to generate realtime accompaniment to previously unknown songs, it must predict their harmonic development, i.e. the chords to be played. We claim that such a system must combine long-term experience, to identify typical chord sequences (e.g. II–V and II–V–I), with ‘on-the-fly’ adaptation to track-recurrent structures (e.g. choruses and refrains) of the particular song being played. We have implemented a prediction system using a neural network model that encompasses prior knowledge about typical chord sequences. The results achieved are very encouraging, and rather better than those reported in the literature. However, our predictor could not adapt its behaviour to the idiosyncrasies of each song, since online learning is difficult in neural networks. In this paper, we propose an extension to our previous work by the inclusion of a rule-based sequence tracker, which detects recurrent chord sequences while the song is being performed. We show that this hybrid model, which combines a neural network predictor with a rule-based sequence tracker, improves the system's performance.
Creating distinctive harmonizations in an identifiable style may be one of the most difficult tasks for amateur song writers, a novel and acceptable melody being relatively easier to produce; and this difficulty may result in the abandonment of otherwise worthwhile projects. To model and assist in this creative process, we propose a hy-brid system for generating style-specific accompaniment, which is capable of creating new harmonizations for melodies, with proper harmonic resolutions, in a style that is learned from only a few examples. In the proposed system, a chord tone determination module first learns, then determines, which notes in a given melody are likely chord tones. According to these chord tones, triads are assigned first to the bars with unambiguous solutions, and these triads serve as checkpoints. The system then con-structs possible chord progressions using neo-Riemannian transforms between checkpoints, and represents the al-ternate paths in a tree structure. A Markov chain with learned probabilities for these neo-Riemanian transforms then generates the final chord progression. We select four songs by the British rock band, Radiohead, to evaluate the system. Three songs are used for training, and an accom-paniment is generated for the held out melody. We present the results of two case studies. We find that the system generates chords closely related to the original, and the resulting chord transitions reinforce the phrase structure of the melody.
We present a method for generating harmonic progressions using case-based analysis of existing material that employs a Markov model. Using a unique method for specifying desired harmonic complexity, tension between chord transitions, and a desired bass-line, the user specifies a 3 dimensional vector, which the realtime generative algorithm attempts to match during chord sequence generation. The proposed system thus offers a balance between user-requested material and coherence within the database.
Many computational models of music fail to capture essential aspects of the high-level musical structure and context, and this limits their usefulness, particularly for musically informed users. We describe two recent approaches to modelling musical harmony, using a probabilistic and a logic-based framework respectively, which attempt to reduce the gap between computational models and human understanding of music. The first is a chord transcription system which uses a high-level model of musical context in which chord, key, metrical position, bass note, chroma features and repetition structure are integrated in a Bayesian framework, achieving state-of-the-art performance. The second approach uses inductive logic programming to learn logical descriptions of harmonic sequences which characterise particular styles or genres. Each approach brings us one step closer to modelling music in the way it is conceptualised by musicians.
This paper introduces decision-theoretic planning techniques into automatic mu- sic generation. Markov decision processes (MDPs) are a mathematical model of plan- ning under uncertainty, and factored MDPs demonstrate great advantages in an environ- ment where the state space is large. Given a melody, we use a factored MDP planner to fill in the other three voices, based on the characteristics of classical Western four-part harmony.
In this paper we propose a text represention for musical chord symbols that is simple and intuitive for musically trained individuals to write and understand, yet highly structured and unambiguous to parse with computer pro- grams. When designing feature extraction algorithms, it is important to have a hand annotated test set providing a ground truth to compare results against. Hand labelling of chords in music les is a long and arduous task and there is no standard annotation methodology, which causes dif- culties sharing with existing annotations. In this pa- per we address this problem by dening a rigid, context- independent syntax for representing chord symbols in text, supported with a new database of annotations using this system.
Max Mathews was last interviewed for Computer Music Journal in 1980 in an article by Curtis Roads. The present interview took place at Max Mathews's home in San Francisco, California, in late May 2008. (See Figure 1.) This project was an interesting one, as I had the opportunity to stay at his home and conduct the interview, which I video-recorded in HD format over the course of one week. The original set of video recordings lasted approximately three hours total. I then edited them down to a duration of approximately one and one-half hours for inclusion on the 2009 Computer Music Journal Sound and Video Anthology, which will accompany the Winter 2009 issue of the Journal. The following article is a lightly edited transcript of that edited video recording.
Park: Could you tell me a little bit about your background: where you grew up, where you went to school—a little bit about your background that we don't usually hear in interviews.
Mathews: Oh, I'd be glad to. I was born and grew up in the middle of the country in Nebraska. My parents were both schoolteachers. They both really liked teaching sciences. My father taught physics, chemistry, and biology in high school and was also the principal of the high school. It was a small school, with class sizes of about twelve students, and it was a very good place to begin an education. My father let me play in the physics, biology, and chemistry laboratories, so I enjoyed making lots of things—making motors that would run, making barometers out of mercury, playing with mercury—you could do that in those days.
Park: Hopefully you didn't hold it in your hands.
Mathews: Oh yes, I held it in my hands, and I am still here at 80. One of the important things I learned in school was how to touch-type; that has become very useful now that computers have come along. I also was taught in the ninth grade how to study by myself. That is when students were introduced to algebra. Most of the farmers and their sons in the area didn't care about learning algebra, and they didn't need it in their work. So, the math teacher gave me a book and I and two or three other students worked the problems in the book and learned algebra for ourselves. And this was such a wonderful way of learning that after I finished the algebra book, I got a calculus book and spent the next few years learning calculus by myself. I never really graduated from high school; I just stopped going there.
This was in 1944, and instead I took an exam for the Navy and enlisted as a radar repairman and essentially fell in love with electronics at that time. [Editor's note: Mr. Mathews moved to Seattle.] I did find out that the two schools that the teachers in Seattle recommended were Cal Tech [California Institute of Technology] and MIT [Massachusetts Institute of Technology]. Since [my wife] Marjorie was in California and was going to Stanford [University], I chose Cal Tech, and that was a very lucky and wise choice, as the undergraduate education I got at Cal Tech was superb. The techniques I learned in math and physics in freshman and sophomore classes at Cal Tech were the techniques that enabled me to pass the doctoral qualifying examinations when I went to MIT for my graduate work. On the other hand, even though I graduated from Cal Tech in electrical and mechanical engineering, when I got out of Cal Tech I didn't know how to simply build a simple audio amplifier; but at MIT I learned how to build complicated computer circuits out of the rudimentary operational amplifiers that were around at that time.
Another great stroke of luck came when I was refused employment at the 3M Company after they had offered me a job, as a result of a back injury that I had, so instead of going to Minneapolis—which I favored, since my family still lived in Nebraska—I went to Bell Telephone Laboratories (Figure...
In this paper we survey the use of different AI methods for algorithmic composition, present their advantages and disadvantages, discuss some important general issues and propose desirable future prospects
The paper is a review article comparing a number of approaches to natural language syntax and semantics that have been developed using categorial frameworks.
It distinguishes two related but distinct varieties of categorial theory, one related to Natural Deduction systems and the axiomatic calculi of Lambek, and another which involves more specialized combinatory operations.
Genetic Algorithms (GA's) are considered promising for music composition because they combine ‘creativity’ (ability to explore a large search space) with constraints (creative 'excess' is 'pruned' using a fitness function). A major difficulty with the use of GA's for this task is to define fitness functions which capture the aesthetic qualities of the wide range of successful melodies. In this paper we report on research that addresses this problem in the context of a modest compositional task, melodic extension. We describe 21 melodic features used as the basis for a GA fitness function and for mutation procedures. We discuss how the features were chosen, measured for significance, and might be incorporated into a fitness function.
This paper describes a system for the generation of jazz melodies over an input chord progression. A genetic algorithm was used to search through the space of possible solutions. A symbolic, as opposed to binary, approach with domain-specific reproduction operators was chosen because it allowed knowledge based constraints to be imposed on the search space. The objective, algorithmic fitness function as well as the domain-specific genetic operators orientate the search to promising musical paths.
. There are two distinct types of creativity: the flash out of the blue (inspiration ? genius?), and the process of incremental revisions (hard work). Not only are we years away from modeling the former, we do not even begin to understand it. The latter is algorithmic in nature and has been modeled in many systems both musical and non-musical. Algorithmic composition is as old as music composition. It is often considered a cheat, a way out when the composer needs material and/or inspiration. It can also be thought of as a compositional tool that simply makes the composer's work go faster. This article makes a case for algorithmic composition as such a tool. The `hard work' type of creativity often involves trying many different combinations against each other and choosing one over others. This iterative task seems natural to be expressed as a computer algorithm. The implementation issues can be reduced to two components: how to understand one's own creative process well enough to repro...
Presented is an application of genetic algorithms to the problem of composing music, in which GAs are used to produce a set of data filters that identify acceptable material from the output of a stochastic music generator. The algorithmic composition system variations is described and musical examples of its output are given. Also discussed briefly is the system's application to microtonal music.
Problem solving can be compared to a search in a state space. The description of the domain knowledge must be encoded as a knowledge base, then the problem solver searches through the state space for an appropriate answer. The experiment reported here is a comparison between GAs and a rule-based system searching the same space: that of four-part harmonisations of chorale melodies. The experiment leads to the conclusion that the quality of the solution fundamentally depends on the knowledge the system possesses and that the KBS is better suited for this task. We then outline an alternative approach of explicitly structuring control knowledge to solve this problem. 1
This paper aims to present a systematic literature review of research in music information seeking and its application to musical creativity and creative activities and in particular composition, performance and improvisation, and listening and analysis. A seed set of 901 articles published between 1973 and 2015 was evaluated and in total 65 studies were considered for further analyses. Data extraction and synthesis was performed through content analysis using the PRISMA method. Three thematic categories were identified in regard to music information needs: (a) those related to scholarly activities, (b) musically motivated, as well as (c) those which are related to socializing and communication. In addition, 3 categories of music information sources were connected to musical creativity: (i) those that are related to Internet and media technologies, (ii) those that are related to music libraries, organizations, and music stores, and (iii) those that are related to the subjects' social settings. The paper provides a systematic review, with the aim of showcasing the effect of modern information retrieval techniques in a creative and intensive area of information-dependent activity such as music making and consumption.
We present an automated system for the harmonization of four-part chorales in the style of J.S. Bach. The task of harmonization is divided into two parts, namely (1) generation of a harmonic pro-gression and (2) realization of that progression into melodic lines for the four voice parts. We model the composition of a chord progression with a Hidden Markov Model, and derive a realization of the harmony by solving a constraint satisfaction problem. Our implementation can successfully produce well-formed chorale harmonizations for a variety of melodies.
HarmTrace has shown to be an effective model to derive harmonic functions of chords in their tonal con-text. This article describes the addition of a chord genera-tion and selection mechanism that investigates the abilities of the HarmTrace model as an automatic harmonizer that generates chords with a given melody. A system which is able to generate multiple harmonically well-formed chord sequences for a given melody is added to the model. From the generated sequences the best one is chosen, by means of smallest amount of parser errors. One experiment is carried out in which chord sequences for carefully selected melodies are generated. Subsequently, in a survey a panel of harmony experts are asked, based on listening to the model generated sequences, to describe their professional opinion and rating of these chord sequences. A second ex-periment is carried out in which a chord sequence for an selected melody is generated, and compared to the original accompaniment of that melody that is considered a ground truth. Unlike all currently existing systems, this system guarantees the generation of well-formed chord sequences, in which the constituents are automatically functionally ex-plained with regard to their context. From the experiments, it is shown that the system is capable of generating correct chords per melody tone. However, it is also shown that harmonizing a melody with individually well-formed chord sequences from the HarmTrace grammar do not guarantee a harmonically well-sounding coherence between the sequence and the melody. This can be explained by the nature of the system, which built to analyze, rather than to create music.
Algorithmic composition is the partial or total automation of the process of
music composition by using computers. Since the 1950s, different computational
techniques related to Artificial Intelligence have been used for algorithmic
composition, including grammatical representations, probabilistic methods,
neural networks, symbolic rule-based systems, constraint programming and
evolutionary algorithms. This survey aims to be a comprehensive account of
research on algorithmic composition, presenting a thorough view of the field
for researchers in Artificial Intelligence.
This paper reprots on a rule-based expert system called CHORAL, for harmonization and Schenkerian analysis of chorales in the style of J. S. Bach. The author first briefly compares his approach with some current trends in algorithmic composition and music analysis, and then describes the CHORAL system itself. 35 Refs.
The recursive character of musical chord sequences makes generative grammar a suitable formalism for describing the rules that constrain such sequences. A small number of rules are presented which generate the members of a large class of complex chord sequences that are generally recognised to be closely related, namely the set of jazz 12-bar blues. The rules are illustrated using a testing corpus of jazz chord sequences, and certain extensions are considered.
An abstract is not available.
EMI was founded in 1983 as a project for understanding musical style. The motive for establishing it was purely selfish: the author's desire to have help in finishing a commission. He focused on how to imitate musical styles, since nothing useful could be accomplished if the computer did not understand how to produce viable music in his style. One of the components of emulating style discovered during the years of research since EMI's inception includes non‐linear composition, the focal point of this article. Results of research have been presented at the International Computer Music Conference (1987, Champaign, IL) and its resultant proceedings, at the International Symposium: Charles Ives and the American Music Tradition up to the Present (Cologne, Germany, February, 1988), the American Association of Artificial Intelligence (St. Paul, August, 1988), AIM, First International Workshop on Al and Music (Bonn, Germany, September, 1988), as well as articles in the Computer Music Journal (Winter, 1987) and Al Expert Magazine (March, 1988). Studies in the replication of musical styles have revealed that top‐down approaches to music creation produce more elegant results than linear or note‐by‐note processes. As well, object‐orientation provides an apropos venue for computer realization of non‐linear techniques. EMI uses linguistic protocols of motive and cadence properties which restrict note production. At upper levels, however, the direction of composition is decided by complex interactions between phrases, sections and even movements of a work, all nested in the basic fabric of an initial provocation (of the composition itself). Non‐linear composition has allowed creation of logical large‐scale musical forms not otherwise available from more local systems of procreating detail oriented surface strata. Since this article covers only a small portion of the EMI system, an appendix is provided to demonstrate how non‐linear techniques fit into the larger EMI algorithm.
We describe how we used a data set of chorale harmonisations composed by Johann Sebastian Bach to train Hidden Markov Models. Using a prob- abilistic framework allows us to create a harmonisation system which learns from examples, and which can compose new harmonisations. We make a quantitative comparison of our system's harmonisati on perfor- mance against simpler models, and provide example harmonisations.
Researchers wishing to create computational systems that themselves generate artworks face two interacting challenges. The first is that the standards by which artistic output is judged are notoriously difficult to quantify. The larger AI community is currently involved in a rich internal dialogue on methodolog- ical issues, standards, and rigor, and hence murkiness with re- gard to the assessment of output must be faced squarely. The second challenge is that any artwork exists within an extraordi- narily rich cultural and historical context, and it is rare that an artist who is ignorant of this context will produce acceptable works. In this paper we assert that these considerations argue for case-based AI/Art systems that take critical criteria as pa- rameters. We describe an example system that produces new bebop jazz melodies from a case-base of melodies, using ge- netic programming techniques and a fitness function based on user-provided critical criteria. We discuss the role that such techniques may play in future work on AI and the arts.
We introduce MySong, a system that automatically chooses chords to accompany a vocal melody. A user with no musical experience can create a song with instrumental accompaniment just by singing into a microphone, and can experiment with different styles and chord patterns using interactions designed to be intuitive to non-musicians. We describe the implementation of MySong, which trains a Hidden Markov Model using a music database and uses that model to select chords for new melodies. Model parameters are intuitively exposed to the user. We present results from a study demonstrating that chords assigned to melodies using MySong and chords assigned manually by musicians receive similar subjective ratings. We then present results from a second study showing that thirteen users with no background in music theory are able to rapidly create musical accompaniments using MySong, and that these accompaniments are rated positively by evaluators.
Chord progressions are the building blocks from which tonal music is constructed. Inferring chord progressions is thus an essential step towards modeling long term de- pendencies in music. In this paper, a distributed repre- sentation for chords is designed such that Euclidean dis- tances roughly correspond to psychoacoustic dissimilar- ities. Estimated probabilities of chord substitutions are derived from this representation and are used to introduce smoothing in graphical models observing chord progres- sions. Parameters in the graphical models are learnt with the EM algorithm and the classical Junction Tree algo- rithm is used for inference. Various model architectures are compared in terms of conditional out-of-sample like- lihood. Both perceptual and statistical evidence show that binary trees related to meter are well suited to capture chord dependencies.
Computers in Music Education: Amplifying Musicality addresses the question of how computer technologies might best assist music education. Aimed at current and preservice music teachers and designed to be a basic teaching text, it addresses issues from pedagogy through using computers to aid production and presentation of students' musical works, as well as implementing a computer-aided program in the classroom.
Computing technology has been rapidly changing, particularly when it comes to the field of music. From notation software to MIDI sound creation, from downloading music from the web for study and enjoyment to posting personal MP3s for mass distribution, the world of computers has transformed how we perform and learn about music. This book, written by a music educator and digital media specialist, cuts through the jargon to present a concise, easy-to-digest overview of the field, with practical suggestions spread throughout.
The recent literature on profile hidden Markov model (profile HMM) methods and software is reviewed. Profile HMMs turn a multiple
sequence alignment into a position-specific scoring system suitable for searching databases for remotely homologous sequences.
Profile HMM analyses complement standard pairwise comparison methods for large-scale sequence analysis. Several software implementations
and two large libraries of profile HMMs of common protein domains are available. HMM methods performed comparably to threading
methods in the CASP2 structure prediction exercise.
dren, to compose music is a difficult task. We can view the problem as a spectrum of tasks that range from the development of musical algorithms for automating the compositional process to designing an appropriate interface for humans to interact with the machine. The Hyperscore software tool attempts to address both of these issues. As a graphical environment that facilitates composition through intelligently mapping musical fea-tures to graphical abstractions, Hyperscore provides a visual ana-logue for what is happening struc-turally in the music as opposed to displaying musical events in proce-dural notation or as a set of para-meters, as is often the case with other graphical composition sys-tems. The fundamental idea of Hyperscore is that anyone can per-form two key creative activities without musical training: compose short melodies and describe the large-scale shape of a piece. Providing graphical means to engage in these two activities forms the basis for Hyperscore's functionality. There have been numerous past examples of graphi-cal computer-assisted composition systems. Many of them are suited for professional musicians, using graph-ical objects to represent musical functions or tweak para-meters. The latter category includes "traditional" commercial applications such as Digital Performer, Cubase, and Vision, multitrack sequencers that use graphical input to manipulate low-level parameters. The former category contains PatchWork/OpenMusic,, an object-oriented environment designed by researchers at the Institut de Recherche et Coordination Acoustique/ Musique (IRCAM) to encapsulate musical functions in graphical objects that can be dragged, dropped, and interconnected to implement musical algorithms. Less
The Viterbi algorithm (VA) is a recursive optimal solution to the problem of estimating the state sequence of a discrete-time finite-state Markov process observed in memoryless noise. Many problems in areas such as digital communications can be cast in this form. This paper gives a tutorial exposition of the algorithm and of how it is implemented and analyzed. Applications to date are reviewed. Increasing use of the algorithm in a widening variety of areas is foreseen.
A technique for harmonic analysis is presented that partitions a piece of music into contiguous regions and labels each with the key, mode, and functional chord, e.g. tonic, dominant, etc. The analysis is performed with a hidden Markov model and, as such, is automatically trainable from generic MIDI files and capable of finding the globally optimal harmonic labeling.