Vector quantization (VQ) is a technique to deterministically learn features with discrete codebook representations. It is commonly performed with a variational autoencoding model, VQ-VAE, which can be further extended to hierarchical structures for making high-fidelity reconstructions. However, such hierarchical extensions of VQ-VAE often suffer from the codebook/layer collapse issue, where the codebook is not efficiently used to express the data, and hence degrades reconstruction accuracy. To mitigate this problem, we propose a novel unified framework to stochastically learn hierarchical discrete representation on the basis of the variational Bayes framework, called hierarchically quantized variational autoencoder (HQ-VAE). HQ-VAE naturally generalizes the hierarchical variants of VQ-VAE, such as VQ-VAE-2 and residual-quantized VAE (RQ-VAE), and provides them with a Bayesian training scheme. Our comprehensive experiments on image datasets show that HQ-VAE enhances codebook usage and improves reconstruction performance. We also validated HQ-VAE in terms of its applicability to a different modality with an audio dataset.
In recent years, research on music transcription has focused mainly on architecture design and instrument-specific data acquisition. With the lack of availability of diverse datasets, progress is often limited to solo-instrument tasks such as piano transcription. Several works have explored multi-instrument transcription as a means to bolster the performance of models on low-resource tasks, but these methods face the same data availability issues. We propose Timbre-Trap, a novel framework which unifies music transcription and audio reconstruction by exploiting the strong separability between pitch and timbre. We train a single U-Net to simultaneously estimate pitch salience and reconstruct complex spectral coefficients, selecting between either output during the decoding stage via a simple switch mechanism. In this way, the model learns to produce coefficients corresponding to timbre-less audio, which can be interpreted as pitch salience. We demonstrate that the framework leads to performance comparable to state-of-the-art instrument-agnostic transcription methods, while only requiring a small amount of annotated data.
This paper summarizes the music demixing (MDX) track of the Sound Demixing Challenge (SDX'23). We provide a summary of the challenge setup and introduce the task of robust music source separation (MSS), i.e., training MSS models in the presence of errors in the training data. We propose a formalization of the errors that can occur in the design of a training dataset for MSS systems and introduce two new datasets that simulate such errors: SDXDB23_LabelNoise and SDXDB23_Bleeding1. We describe the methods that achieved the highest scores in the competition. Moreover, we present a direct comparison with the previous edition of the challenge (the Music Demixing Challenge 2021): the best performing system under the standard MSS formulation achieved an improvement of over 1.6dB in signal-to-distortion ratio over the winner of the previous competition, when evaluated on MDXDB21. Besides relying on the signal-to-distortion ratio as objective metric, we also performed a listening test with renowned producers/musicians to study the perceptual quality of the systems and report here the results. Finally, we provide our insights into the organization of the competition and our prospects for future editions.
Speech super-resolution (SR) is a task to increase speech sampling rate by generating high-frequency components. Existing speech SR methods are trained in constrained experimental settings, such as a fixed upsampling ratio. These strong constraints can potentially lead to poor generalization ability in mismatched real-world cases. In this paper, we propose a neural vocoder based speech super-resolution method (NVSR) that can handle a variety of input resolution and upsampling ratios. NVSR consists of a mel-bandwidth extension module, a neural vocoder module, and a post-processing module. Our proposed system achieves state-of-the-art results on the VCTK multi-speaker benchmark. On 44.1 kHz target resolution, NVSR outperforms WSRGlow and Nu-wave by 8% and 37% respectively on log spectral distance and achieves a significantly better perceptual quality. We also demonstrate that prior knowledge in the pre-trained vocoder is crucial for speech SR by performing mel-bandwidth extension with a simple replication-padding method. Samples can be found in https://haoheliu.github.io/nvsr.
In recent years, neural network based methods have been proposed as a method that cangenerate representations from music, but they are not human readable and hardly analyzable oreditable by a human. To address this issue, we propose a novel method to learn source-awarelatent representations of music through Vector-Quantized Variational Auto-Encoder(VQ-VAE).We train our VQ-VAE to encode an input mixture into a tensor of integers in a discrete latentspace, and design them to have a decomposed structure which allows humans to manipulatethe latent vector in a source-aware manner. This paper also shows that we can generate basslines by estimating latent vectors in a discrete space.
Conditioned source separations have attracted significant attention because of their flexibility, applicability and extensionality. Their performance was usually inferior to the existing approaches, such as the single source separation model. However, a recently proposed method called LaSAFT-Net has shown that conditioned models can show comparable performance against existing single-source separation models. This paper presents LightSAFT-Net, a lightweight version of LaSAFT-Net. As a baseline, it provided a sufficient SDR performance for comparison during the Music Demixing Challenge at ISMIR 2021. This paper also enhances the existing LightSAFT-Net by replacing the LightSAFT blocks in the encoder with TFC-TDF blocks. Our enhanced LightSAFT-Net outperforms the previous one with fewer parameters.
Recently, many methods based on deep learning have been proposed for music source separation. Some state-of-the-art methods have shown that stacking many layers with many skip connections improve the SDR performance. Although such a deep and complex architecture shows outstanding performance, it usually requires numerous computing resources and time for training and evaluation. This paper proposes a two-stream neural network for music demixing, called KUIELab-MDX-Net, which shows a good balance of performance and required resources. The proposed model has a time-frequency branch and a time-domain branch, where each branch separates stems, respectively. It blends results from two streams to generate the final estimation. KUIELab-MDX-Net took second place on leaderboard A and third place on leaderboard B in the Music Demixing Challenge at ISMIR 2021. This paper also summarizes experimental results on another benchmark, MUSDB18. Our source code is available online.
This paper proposes a neural network that performs audio transformations to user-specified sources (e.g., vocals) of a given audio track according to a given description while preserving other sources not mentioned in the description. Audio Manipulation on a Specific Source (AMSS) is challenging because a sound object (i.e., a waveform sample or frequency bin) is `transparent'; it usually carries information from multiple sources, in contrast to a pixel in an image. To address this challenging problem, we propose AMSS-Net, which extracts latent sources and selectively manipulates them while preserving irrelevant sources. We also propose an evaluation benchmark for several AMSS tasks, and we show that AMSS-Net outperforms baselines on several AMSS tasks via objective metrics and empirical verification.
Recent deep-learning approaches have shown that Frequency Transformation (FT) blocks can significantly improve spectrogram-based single-source separation models by capturing frequency patterns. The goal of this paper is to extend the FT block to fit the multi-source task. We propose the Latent Source Attentive Frequency Transformation (LaSAFT) block to capture source-dependent frequency patterns. We also propose the Gated Point-wise Convolutional Modulation (GPoCM), an extension of Feature-wise Linear Modulation (FiLM), to modulate internal features. By employing these two novel methods, we extend the Conditioned-U-Net (CUNet) for multi-source separation, and the experimental results indicate that our LaSAFT and GPoCM can improve the CUNet's performance, achieving state-of-the-art SDR performance on several MUSDB18 source separation tasks.
Musical Source Separation (MSS) is a signal processing task that tries to separate the mixed musical signal into each acoustic sound source, such as singing voice or drums. Recently many machine learning-based methods have been proposed for the MSS task, but there were no existing works that evaluate and directly compare various types of networks. In this paper, we aim to design a variety of neural transformation methods, including time-invariant methods, time-frequency methods, and mixtures of two different transformations. Our experiments provide abundant material for future works by comparing several transformation methods. We train our models on raw complex-valued STFT outputs and achieve state-of-the-art SDR performance on the MUSDB singing voice separation task by a large margin of 1.0 dB.