Generating music has a few notable differences from generating images and videos. First, music is an art of time, necessitating a temporal model. Second, music is usually composed of multiple instruments/tracks with their own temporal dynamics, but collectively they unfold over time interdependently. Lastly, musical notes are often grouped into chords, arpeggios or melodies in polyphonic music, and thereby introducing a chronological ordering of notes is not naturally suitable. In this paper, we propose three models for symbolic multi-track music generation under the framework of generative adversarial networks (GANs). The three models, which differ in the underlying assumptions and accordingly the network architectures, are referred to as the jamming model, the composer model and the hybrid model. We trained the proposed models on a dataset of over one hundred thousand bars of rock music and applied them to generate piano-rolls of five tracks: bass, drums, guitar, piano and strings. A few intra-track and inter-track objective metrics are also proposed to evaluate the generative results, in addition to a subjective user study. We show that our models can generate coherent music of four bars right from scratch (i.e. without human inputs). We also extend our models to human-AI cooperative music generation: given a specific track composed by human, we can generate four additional tracks to accompany it. All code, the dataset and the rendered audio samples are available at https://salu133445.github.io/musegan/ .

We propose a novel approach for the generation of polyphonic music based on LSTMs. We generate music in two steps. First, a chord LSTM predicts a chord progression based on a chord embedding. A second LSTM then generates polyphonic music from the predicted chord progression. The generated music sounds pleasing and harmonic, with only few dissonant notes. It has clear long-term structure that is similar to what a musician would play during a jam session. We show that our approach is sensible from a music theory perspective by evaluating the learned chord embeddings. Surprisingly, our simple model managed to extract the circle of fifths, an important tool in music theory, from the dataset.

Currently, almost all the multi-track music generation models use the Convolutional Neural Network (CNN) to build the generative model, while the Recurrent Neural Network (RNN) based models can not be applied in this task. In view of the above problem, this paper proposes a RNN-based Hierarchical Multi-modal Fusion Generation Variational Autoencoder (VAE) network, MIDI-Sandwich2, for multi-track symbolic music generation. Inspired by VQ-VAE2, MIDI-Sandwich2 expands the dimension of the original hierarchical model by using multiple independent Binary Variational Autoencoder (BVAE) models without sharing weights to process the information of each track. Then, with multi-modal fusion technology, the upper layer named Multi-modal Fusion Generation VAE (MFG-VAE) combines the latent space vectors generated by the respective tracks, and uses the decoder to perform the ascending dimension reconstruction to simulate the inverse operation of multi-modal fusion, multi-modal generation, so as to realize the RNN-based multi-track symbolic music generation. For the multi-track format pianoroll, we also improve the output binarization method of MuseGAN, which solves the problem that the refinement step of the original scheme is difficult to differentiate and the gradient is hard to descent, making the generated song more expressive. The model is validated on the Lakh Pianoroll Dataset (LPD) multi-track dataset. Compared to the MuseGAN, MIDI-Sandwich2 can not only generate harmonious multi-track music, the generation quality is also close to the state of the art level. At the same time, by using the VAE to restore songs, the semi-generated songs reproduced by the MIDI-Sandwich2 are more beautiful than the pure autogeneration music generated by MuseGAN. Both the code and the audition audio samples are open source on https://github.com/LiangHsia/MIDI-S2.

Most existing neural network models for music generation explore how to generate music bars, then directly splice the music bars into a song. However, these methods do not explore the relationship between the bars, and the connected song as a whole has no musical form structure and sense of musical direction. To address this issue, we propose a Multi-model Multi-task Hierarchical Conditional VAE-GAN (Variational Autoencoder-Generative adversarial networks) networks, named MIDI-Sandwich, which combines musical knowledge, such as musical form, tonic, and melodic motion. The MIDI-Sandwich has two submodels: Hierarchical Conditional Variational Autoencoder (HCVAE) and Hierarchical Conditional Generative Adversarial Network (HCGAN). The HCVAE uses hierarchical structure. The underlying layer of HCVAE uses Local Conditional Variational Autoencoder (L-CVAE) to generate a music bar which is pre-specified by the First and Last Notes (FLN). The upper layer of HCVAE uses Global Variational Autoencoder(G-VAE) to analyze the latent vector sequence generated by the L-CVAE encoder, to explore the musical relationship between the bars, and to produce the song pieced together by multiple music bars generated by the L-CVAE decoder, which makes the song both have musical structure and sense of direction. At the same time, the HCVAE shares a part of itself with the HCGAN to further improve the performance of the generated music. The MIDI-Sandwich is validated on the Nottingham dataset and is able to generate a single-track melody sequence (17x8 beats), which is superior to the length of most of the generated models (8 to 32 beats). Meanwhile, by referring to the experimental methods of many classical kinds of literature, the quality evaluation of the generated music is performed. The above experiments prove the validity of the model.

Music generation research has grown in popularity over the past decade, thanks to the deep learning revolution that has redefined the landscape of artificial intelligence. In this paper, we propose a novel approach to music generation inspired by musical segment concatenation methods and hash learning algorithms. Given a segment of music, we use a deep recurrent neural network and ranking-based hash learning to assign a forward hash code to the segment to retrieve candidate segments for continuation with matching backward hash codes. The proposed method is thus called Deep Segment Hash Learning (DSHL). To the best of our knowledge, DSHL is the first end-to-end segment hash learning method for music generation, and the first to use pair-wise training with segments of music. We demonstrate that this method is capable of generating music which is both original and enjoyable, and that DSHL offers a promising new direction for music generation research.

This paper presents a deep reinforcement learning algorithm for online accompaniment generation, with potential for real-time interactive human-machine duet improvisation. Different from offline music generation and harmonization, online music accompaniment requires the algorithm to respond to human input and generate the machine counterpart in a sequential order. We cast this as a reinforcement learning problem, where the generation agent learns a policy to generate a musical note (action) based on previously generated context (state). The key of this algorithm is the well-functioning reward model. Instead of defining it using music composition rules, we learn this model from monophonic and polyphonic training data. This model considers the compatibility of the machine-generated note with both the machine-generated context and the human-generated context. Experiments show that this algorithm is able to respond to the human part and generate a melodic, harmonic and diverse machine part. Subjective evaluations on preferences show that the proposed algorithm generates music pieces of higher quality than the baseline method.

In this paper AlimNet (With respect to great musician, Alim Qasimov) an auxiliary generative adversarial deep neural network (ACGAN) for generating music categorically, is used. This proposed network is a conditional ACGAN to condition the generation process on music tracks which has a hybrid architecture, composing of different kind of layers of neural networks. The employed music dataset is MICM which contains 1137 music samples (506 violins and 631 straw) with seven types of classical music Dastgah labels. To extract both temporal and spectral features, Short-Time Fourier Transform (STFT) is applied to convert input audio signals from time domain to time-frequency domain. GANs are composed of a generator for generating new samples and a discriminator to help generator making better samples. Samples in time-frequency domain are used to train discriminator in fourteen classes (seven Dastgahs and two instruments). The outputs of the conditional ACGAN are also artificial music samples in those mentioned scales in time-frequency domain. Then the output of the generator is transformed by Inverse STFT (ISTFT). Finally, randomly ten generated music samples (five violin and five straw samples) are given to ten musicians to rate how exact the samples are and the overall result was 76.5%.

Music Generation (MG) is an interesting research topic that links the art of music and Artificial Intelligence (AI). The goal is to train an artificial composer to generate infinite, fresh, and pleasurable musical pieces. Music has different parts such as melody, harmony, and rhythm. In this paper, we propose a Generative Genetic Algorithm (GGA) to produce a melody automatically. The main GGA uses a Long Short-Term Memory (LSTM) recurrent neural network as the objective function, which should be trained by a spectrum of bad-to-good melodies. These melodies have to be provided by another GGA with a different objective function. Good melodies have been provided by CAMPINs collection. We have considered the rhythm in this work, too. The experimental results clearly show that the proposed GGA method is able to generate eligible melodies with natural transitions and without rhythm error.

In this paper we introduce a novel feature augmentation approach for generating structured musical compositions comprising melodies and harmonies. The proposed method augments a connectionist generation model with count-down to song conclusion and meter markers as extra input features to study whether neural networks can learn to produce more aesthetically pleasing and structured musical output as a consequence of augmenting the input data with structural features. An RNN architecture with LSTM cells is trained on the Nottingham folk music dataset in a supervised sequence learning setup, following a Music Language Modelling approach, and then applied to generation of harmonies and melodies. Our experiments show an improved prediction performance for both types of annotation. The generated music was also subjectively evaluated using an on-line Turing style listening test which confirms a substantial improvement in the aesthetic quality and in the perceived structure of the music generated using the temporal structure.

With recent breakthroughs in artificial neural networks, deep generative models have become one of the leading techniques for computational creativity. Despite very promising progress on image and short sequence generation, symbolic music generation remains a challenging problem since the structure of compositions are usually complicated. In this study, we attempt to solve the melody generation problem constrained by the given chord progression. This music meta-creation problem can also be incorporated into a plan recognition system with user inputs and predictive structural outputs. In particular, we explore the effect of explicit architectural encoding of musical structure via comparing two sequential generative models: LSTM (a type of RNN) and WaveNet (dilated temporal-CNN). As far as we know, this is the first study of applying WaveNet to symbolic music generation, as well as the first systematic comparison between temporal-CNN and RNN for music generation. We conduct a survey for evaluation in our generations and implemented Variable Markov Oracle in music pattern discovery. Experimental results show that to encode structure more explicitly using a stack of dilated convolution layers improved the performance significantly, and a global encoding of underlying chord progression into the generation procedure gains even more.

In addition to traditional tasks such as prediction, classification and translation, deep learning is receiving growing attention as an approach for music generation, as witnessed by recent research groups such as Magenta at Google and CTRL (Creator Technology Research Lab) at Spotify. The motivation is in using the capacity of deep learning architectures and training techniques to automatically learn musical styles from arbitrary musical corpora and then to generate samples from the estimated distribution. However, a direct application of deep learning to generate content rapidly reaches limits as the generated content tends to mimic the training set without exhibiting true creativity. Moreover, deep learning architectures do not offer direct ways for controlling generation (e.g., imposing some tonality or other arbitrary constraints). Furthermore, deep learning architectures alone are autistic automata which generate music autonomously without human user interaction, far from the objective of interactively assisting musicians to compose and refine music. Issues such as: control, structure, creativity and interactivity are the focus of our analysis. In this paper, we select some limitations of a direct application of deep learning to music generation, analyze why the issues are not fulfilled and how to address them by possible approaches. Various examples of recent systems are cited as examples of promising directions.

Most existing neural network models for music generation use recurrent neural networks. However, the recent WaveNet model proposed by DeepMind shows that convolutional neural networks (CNNs) can also generate realistic musical waveforms in the audio domain. Following this light, we investigate using CNNs for generating melody (a series of MIDI notes) one bar after another in the symbolic domain. In addition to the generator, we use a discriminator to learn the distributions of melodies, making it a generative adversarial network (GAN). Moreover, we propose a novel conditional mechanism to exploit available prior knowledge, so that the model can generate melodies either from scratch, by following a chord sequence, or by conditioning on the melody of previous bars (e.g. a priming melody), among other possibilities. The resulting model, named MidiNet, can be expanded to generate music with multiple MIDI channels (i.e. tracks). We conduct a user study to compare the melody of eight-bar long generated by MidiNet and by Google's MelodyRNN models, each time using the same priming melody. Result shows that MidiNet performs comparably with MelodyRNN models in being realistic and pleasant to listen to, yet MidiNet's melodies are reported to be much more interesting.

Existing automatic music generation approaches that feature deep learning can be broadly classified into two types: raw audio models and symbolic models. Symbolic models, which train and generate at the note level, are currently the more prevalent approach; these models can capture long-range dependencies of melodic structure, but fail to grasp the nuances and richness of raw audio generations. Raw audio models, such as DeepMind's WaveNet, train directly on sampled audio waveforms, allowing them to produce realistic-sounding, albeit unstructured music. In this paper, we propose an automatic music generation methodology combining both of these approaches to create structured, realistic-sounding compositions. We consider a Long Short Term Memory network to learn the melodic structure of different styles of music, and then use the unique symbolic generations from this model as a conditioning input to a WaveNet-based raw audio generator, creating a model for automatic, novel music. We then evaluate this approach by showcasing results of this work.

The variational autoencoder (VAE) is a popular probabilistic generative model. However, one shortcoming of VAEs is that the latent variables cannot be discrete, which makes it difficult to generate data from different modes of a distribution. Here, we propose an extension of the VAE framework that incorporates a classifier to infer the discrete class of the modeled data. To model sequential data, we can combine our Classifying VAE with a recurrent neural network such as an LSTM. We apply this model to algorithmic music generation, where our model learns to generate musical sequences in different keys. Most previous work in this area avoids modeling key by transposing data into only one or two keys, as opposed to the 10+ different keys in the original music. We show that our Classifying VAE and Classifying VAE+LSTM models outperform the corresponding non-classifying models in generating musical samples that stay in key. This benefit is especially apparent when trained on untransposed music data in the original keys.

The current tsunami of deep learning (the hyper-vitamined return of artificial neural networks) applies not only to traditional statistical machine learning tasks: prediction and classification (e.g., for weather prediction and pattern recognition), but has already conquered other areas, such as translation. A growing area of application is the generation of creative content: in particular the case of music, the topic of this paper. The motivation is in using the capacity of modern deep learning techniques to automatically learn musical styles from arbitrary musical corpora and then to generate musical samples from the estimated distribution, with some degree of control over the generation. This article provides a survey of music generation based on deep learning techniques. After a short introduction to the topic illustrated by a recent exemple, the article analyses some early works from the late 1980s using artificial neural networks for music generation and how their pioneering contributions foreshadowed current techniques. Then, we introduce some conceptual framework to analyze the various concepts and dimensions involved. Various examples of recent systems are introduced and analyzed to illustrate the variety of concerns and of techniques.

Several methods exist for a computer to generate music based on data including Markov chains, recurrent neural networks, recombinancy, and grammars. We explore the use of unit selection and concatenation as a means of generating music using a procedure based on ranking, where, we consider a unit to be a variable length number of measures of music. We first examine whether a unit selection method, that is restricted to a finite size unit library, can be sufficient for encompassing a wide spectrum of music. We do this by developing a deep autoencoder that encodes a musical input and reconstructs the input by selecting from the library. We then describe a generative model that combines a deep structured semantic model (DSSM) with an LSTM to predict the next unit, where units consist of four, two, and one measures of music. We evaluate the generative model using objective metrics including mean rank and accuracy and with a subjective listening test in which expert musicians are asked to complete a forced-choiced ranking task. We compare our model to a note-level generative baseline that consists of a stacked LSTM trained to predict forward by one note.

Realistic music generation is a challenging task. When building generative models of music that are learnt from data, typically high-level representations such as scores or MIDI are used that abstract away the idiosyncrasies of a particular performance. But these nuances are very important for our perception of musicality and realism, so in this work we embark on modelling music in the raw audio domain. It has been shown that autoregressive models excel at generating raw audio waveforms of speech, but when applied to music, we find them biased towards capturing local signal structure at the expense of modelling long-range correlations. This is problematic because music exhibits structure at many different timescales. In this work, we explore autoregressive discrete autoencoders (ADAs) as a means to enable autoregressive models to capture long-range correlations in waveforms. We find that they allow us to unconditionally generate piano music directly in the raw audio domain, which shows stylistic consistency across tens of seconds.

Recent neural waveform synthesizers such as WaveNet, WaveGlow, and the neural-source-filter (NSF) model have shown good performance in speech synthesis despite their different methods of waveform generation. The similarity between speech and music audio synthesis techniques suggests interesting avenues to explore in terms of the best way to apply speech synthesizers in the music domain. This work compares three neural synthesizers used for musical instrument sounds generation under three scenarios: training from scratch on music data, zero-shot learning from the speech domain, and fine-tuning-based adaptation from the speech to the music domain. The results of a large-scale perceptual test demonstrated that the performance of three synthesizers improved when they were pre-trained on speech data and fine-tuned on music data, which indicates the usefulness of knowledge from speech data for music audio generation. Among the synthesizers, WaveGlow showed the best potential in zero-shot learning while NSF performed best in the other scenarios and could generate samples that were perceptually close to natural audio.