Recent work has proposed various adversarial losses for training generative adversarial networks. Yet, it remains unclear what certain types of functions are valid adversarial loss functions, and how these loss functions perform against one another. In this paper, we aim to gain a deeper understanding of adversarial losses by decoupling the effects of their component functions and regularization terms. We first derive some necessary and sufficient conditions of the component functions such that the adversarial loss is a divergence-like measure between the data and the model distributions. In order to systematically compare different adversarial losses, we then propose DANTest, a new, simple framework based on discriminative adversarial networks. With this framework, we evaluate an extensive set of adversarial losses by combining different component functions and regularization approaches. This study leads to some new insights into the adversarial losses. For reproducibility, all source code is available at https://github.com/salu133445/dan .
We propose the BinaryGAN, a novel generative adversarial network (GAN) that uses binary neurons at the output layer of the generator. We employ the sigmoid-adjusted straight-through estimators to estimate the gradients for the binary neurons and train the whole network by end-to-end backpropogation. The proposed model is able to directly generate binary-valued predictions at test time. We implement such a model to generate binarized MNIST digits and experimentally compare the performance for different types of binary neurons, GAN objectives and network architectures. Although the results are still preliminary, we show that it is possible to train a GAN that has binary neurons and that the use of gradient estimators can be a promising direction for modeling discrete distributions with GANs. For reproducibility, the source code is available at https://github.com/salu133445/binarygan .
It has been shown recently that deep convolutional generative adversarial networks (GANs) can learn to generate music in the form of piano-rolls, which represent music by binary-valued time-pitch matrices. However, existing models can only generate real-valued piano-rolls and require further post-processing, such as hard thresholding (HT) or Bernoulli sampling (BS), to obtain the final binary-valued results. In this paper, we study whether we can have a convolutional GAN model that directly creates binary-valued piano-rolls by using binary neurons. Specifically, we propose to append to the generator an additional refiner network, which uses binary neurons at the output layer. The whole network is trained in two stages. Firstly, the generator and the discriminator are pretrained. Then, the refiner network is trained along with the discriminator to learn to binarize the real-valued piano-rolls the pretrained generator creates. Experimental results show that using binary neurons instead of HT or BS indeed leads to better results in a number of objective measures. Moreover, deterministic binary neurons perform better than stochastic ones in both objective measures and a subjective test. The source code, training data and audio examples of the generated results can be found at https://salu133445.github.io/bmusegan/ .
The goal of music highlight extraction is to get a short consecutive segment of a piece of music that provides an effective representation of the whole piece. In a previous work, we introduced an attention-based convolutional recurrent neural network that uses music emotion classification as a surrogate task for music highlight extraction, for Pop songs. The rationale behind that approach is that the highlight of a song is usually the most emotional part. This paper extends our previous work in the following two aspects. First, methodology-wise we experiment with a new architecture that does not need any recurrent layers, making the training process faster. Moreover, we compare a late-fusion variant and an early-fusion variant to study which one better exploits the attention mechanism. Second, we conduct and report an extensive set of experiments comparing the proposed attention-based methods against a heuristic energy-based method, a structural repetition-based method, and a few other simple feature-based methods for this task. Due to the lack of public-domain labeled data for highlight extraction, following our previous work we use the RWC POP 100-song data set to evaluate how the detected highlights overlap with any chorus sections of the songs. The experiments demonstrate the effectiveness of our methods over competing methods. For reproducibility, we open source the code and pre-trained model at https://github.com/remyhuang/pop-music-highlighter/.
Research on automatic music generation has seen great progress due to the development of deep neural networks. However, the generation of multi-instrument music of arbitrary genres still remains a challenge. Existing research either works on lead sheets or multi-track piano-rolls found in MIDIs, but both musical notations have their limits. In this work, we propose a new task called lead sheet arrangement to avoid such limits. A new recurrent convolutional generative model for the task is proposed, along with three new symbolic-domain harmonic features to facilitate learning from unpaired lead sheets and MIDIs. Our model can generate lead sheets and their arrangements of eight-bar long. Audio samples of the generated result can be found at https://drive.google.com/open?id=1c0FfODTpudmLvuKBbc23VBCgQizY6-Rk
Can we make a famous rap singer like Eminem sing whatever our favorite song? Singing style transfer attempts to make this possible, by replacing the vocal of a song from the source singer to the target singer. This paper presents a method that learns from unpaired data for singing style transfer using generative adversarial networks.
Vector-valued neural learning has emerged as a promising direction in deep learning recently. Traditionally, training data for neural networks (NNs) are formulated as a vector of scalars; however, its performance may not be optimal since associations among adjacent scalars are not modeled. In this paper, we propose a new vector neural architecture called the Arbitrary BIlinear Product Neural Network (ABIPNN), which processes information as vectors in each neuron, and the feedforward projections are defined using arbitrary bilinear products. Such bilinear products can include circular convolution, seven-dimensional vector product, skew circular convolution, reversed- time circular convolution, or other new products not seen in previous work. As a proof-of-concept, we apply our proposed network to multispectral image denoising and singing voice sepa- ration. Experimental results show that ABIPNN gains substantial improvements when compared to conventional NNs, suggesting that associations are learned during training.
Instrument playing is among the most common scenes in music-related videos, which represent nowadays one of the largest sources of online videos. In order to understand the instrument-playing scenes in the videos, it is important to know what instruments are played, when they are played, and where the playing actions occur in the scene. While audio-based recognition of instruments has been widely studied, the visual aspect of the music instrument playing remains largely unaddressed in the literature. One of the main obstacles is the difficulty in collecting annotated data of the action locations for training-based methods. To address this issue, we propose a weakly-supervised framework to find when and where the instruments are played in the videos. We propose to use two auxiliary models, a sound model and an object model, to provide supervisions for training the instrument-playing action model. The sound model provides temporal supervisions, while the object model provides spatial supervisions. They together can simultaneously provide temporal and spatial supervisions. The resulted model only needs to analyze the visual part of a music video to deduce which, when and where instruments are played. We found that the proposed method significantly improves the localization accuracy. We evaluate the result of the proposed method temporally and spatially on a small dataset (totally 5,400 frames) that we manually annotated.
Singing voice separation attempts to separate the vocal and instrumental parts of a music recording, which is a fundamental problem in music information retrieval. Recent work on singing voice separation has shown that the low-rank representation and informed separation approaches are both able to improve separation quality. However, low-rank optimizations are computationally inefficient due to the use of singular value decompositions. Therefore, in this paper, we propose a new linear-time algorithm called informed group-sparse representation, and use it to separate the vocals from music using pitch annotations as side information. Experimental results on the iKala dataset confirm the efficacy of our approach, suggesting that the music accompaniment follows a group-sparse structure given a pre-trained instrumental dictionary. We also show how our work can be easily extended to accommodate multiple dictionaries using the DSD100 dataset.
Informed by recent work on tensor singular value decomposition and circulant algebra matrices, this paper presents a new theoretical bridge that unifies the hypercomplex and tensor-based approaches to singular value decomposition and robust principal component analysis. We begin our work by extending the principal component pursuit to Olariu's polar $n$-complex numbers as well as their bicomplex counterparts. In so doing, we have derived the polar $n$-complex and $n$-bicomplex proximity operators for both the $\ell_1$- and trace-norm regularizers, which can be used by proximal optimization methods such as the alternating direction method of multipliers. Experimental results on two sets of audio data show that our algebraically-informed formulation outperforms tensor robust principal component analysis. We conclude with the message that an informed definition of the trace norm can bridge the gap between the hypercomplex and tensor-based approaches. Our approach can be seen as a general methodology for generating other principal component pursuit algorithms with proper algebraic structures.