Lyrics-Conditioned Neural Melody Generation

Yu, Yi; Harscoët, Florian; Canales, Simon; M, Gurunath Reddy; Tang, Suhua; Jiang, Junjun

doi:10.1007/978-3-030-37734-2_58

Cited by 12 publications

(7 citation statements)

References 2 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…And thinking about the case that there are many 0s in , we add a dropout layer after LSTM layer. Per relevant literatures [18,[30][31][32][33][34][35][36] and our verification, two LSTM layers are sufficient in learning time series relationship. Thus, we use two LSTM layers in both generator and discriminator.…”

Section: Generator Of Lstm-cgan Modelmentioning

confidence: 51%

LSTM-CGAN: Towards Generating Low-Rate DDoS Adversarial Samples for Blockchain-Based Wireless Network Detection Models

Liu

Yin

2021

IEEE Access

View full text Add to dashboard Cite

Low-rate Distributed DoS (LDDoS) attack is a complex large-scale attack behavior with strong time-domain characteristics in blockchain-based wireless network. Blockchain with Machine learning-based models, as promising ways, are taken to detect them and secure wireless network. However, researchers focused on how to improve models' detection performance and work out new blockchain-based protection technologies during the past decades. Due to lack of evolving data, these models and technologies may have poor stability in the face of confrontational samples. To cope with the problem, this paper proposes a novel LSTM-CGAN method to generate high-quality LDDoS adversarial samples for blockchain-based wireless network detection models. In this method, we give a brief feature analysis about LDDoS attack in blockchainbased wireless network and work out its corresponding time series model firstly. And then, we take use of Long Short-Term Memory Networks (LSTM) to learn relationships among sequenced network packages in the same flow. At last, we establish a Condition Generative Adversarial Networks (CGAN) model to use above relationships as specific conditions for generating mimicking behaviors of LDDoS attacks in blockchain-based wireless network. The experimental results show that these generated adversarial samples based on both public and private datasets can cheat the machine learning detection models, and have the similar attack characteristics with the real samples. Consequently, they can be used as blockchain-based wireless network dataset of machine learning classifiers for training to enhance models' stability.

show abstract

Section: Generator Of Lstm-cgan Modelmentioning

confidence: 51%

LSTM-CGAN: Towards Generating Low-Rate DDoS Adversarial Samples for Blockchain-Based Wireless Network Detection Models

Liu

Yin

2021

IEEE Access

View full text Add to dashboard Cite

show abstract

“…They utilize Bayesian for the optimization policy of their GAN. Yi Yu et al [14] developed a lyrics-conditioned AI melody generator. Their system has three components: lyrics encoder network, melody generation network, and MIDI sequence tuner.…”

Section: Melody Pitchmentioning

confidence: 99%

“…DNNs are optimal for keeping long-range temporal structure and temporal dependencies in a musical piece. The state-of-the-art GAN is used for developing nontraditional music composition applications such the generation of melody that confronts to given lyrics [14], or the creation of a deformable timbre synthesis instrument [20]. Counterpoint rules are better represented in a rule-based system although there are rather promising results from utilizing ANNs for this task, specially NMTs [24].…”

Section: Rhythmmentioning

confidence: 99%

See 1 more Smart Citation

Applications of Computational Intelligence in Computer Music Composition

Siphocly

Salem

El-Horabty

2021

IJICIS

View full text Add to dashboard Cite

Engaging computers in composing musical pieces is a challenging and trending field of research. The musical tasks that can be performed or aided by computers' computational powers, are numerous. This paper is concerned with applications of computational intelligence in music composition. Its main objective is to survey various computational intelligence techniques for performing miscellaneous music composition tasks. To achieve this objective, we first define each music composition task, then we discuss the recent applications of each, and the techniques adopted in them. We also highlight the most suitable techniques for performing each task. Our study shows that the most suitable techniques for human composers imitative systems are case-based reasoning and artificial neural networks. It is also shown that Markov models are more suitable for predicting musical notes based on the given previous notes. Genetic algorithms excel in chord progressions generation. Deep neural networks are clever at capturing temporal information of a musical piece. The state-of-the-art generative adversarial networks produce music as close as possible to real compositions. At the end of this study, we shed the light on many future research directions in the field of computer music composition.

show abstract

“…The dataset used for this task is the Lyrics-Melody Dataset created by Yu et al [32]., which contains aligned lyrics-melody pairs extracted and processed from the Lakh full MIDI Dataset [33]. The dataset contains lyrics and melodies for 7,998 songs with the lyrics pre-tokenized at the syllable level.…”

Section: A Data and Input Representationmentioning

confidence: 99%

Melody-Conditioned Lyrics Generation with SeqGANs

Chen

Lerch

2020

Preprint

View full text Add to dashboard Cite

Automatic lyrics generation has received attention from both music and AI communities for years. Early rule-based approaches have -due to increases in computational power and evolution in data-driven models-mostly been replaced with deep-learning-based systems. Many existing approaches, however, either rely heavily on prior knowledge in music and lyrics writing or oversimplify the task by largely discarding melodic information and its relationship with the text. We propose an end-to-end melody-conditioned lyrics generation system based on Sequence Generative Adversarial Networks (SeqGAN), which generates a line of lyrics given the corresponding melody as the input. Furthermore, we investigate the performance of the generator with an additional input condition: the theme or overarching topic of the lyrics to be generated. We show that the input conditions have no negative impact on the evaluation metrics while enabling the network to produce more meaningful results.

show abstract

Lyrics-Conditioned Neural Melody Generation

Cited by 12 publications

References 2 publications

LSTM-CGAN: Towards Generating Low-Rate DDoS Adversarial Samples for Blockchain-Based Wireless Network Detection Models

LSTM-CGAN: Towards Generating Low-Rate DDoS Adversarial Samples for Blockchain-Based Wireless Network Detection Models

Applications of Computational Intelligence in Computer Music Composition

Melody-Conditioned Lyrics Generation with SeqGANs

Contact Info

Product

Resources

About