Kyogu Lee

Chang-Bin Jeon, Hyeongi Moon, Keunwoo Choi et al.

Separation of multiple singing voices into each voice is a rarely studied area in music source separation research. The absence of a benchmark dataset has hindered its progress. In this paper, we present an evaluation dataset and provide baseline studies for multiple singing voices separation. First, we introduce MedleyVox, an evaluation dataset for multiple singing voices separation. We specify the problem definition in this dataset by categorizing it into i) unison, ii) duet, iii) main vs. rest, and iv) N-singing separation. Second, to overcome the absence of existing multi-singing datasets for a training purpose, we present a strategy for construction of multiple singing mixtures using various single-singing datasets. Third, we propose the improved super-resolution network (iSRNet), which greatly enhances initial estimates of separation networks. Jointly trained with the Conv-TasNet and the multi-singing mixture construction strategy, the proposed iSRNet achieved comparable performance to ideal time-frequency masks on duet and unison subsets of MedleyVox. Audio samples, the dataset, and codes are available on our website (https://github.com/jeonchangbin49/MedleyVox).

Kyungsu Kim, Minju Park, Haesun Joung et al.

As digital music production has become mainstream, the selection of appropriate virtual instruments plays a crucial role in determining the quality of music. To search the musical instrument samples or virtual instruments that make one's desired sound, music producers use their ears to listen and compare each instrument sample in their collection, which is time-consuming and inefficient. In this paper, we call this task as Musical Instrument Retrieval and propose a method for retrieving desired musical instruments using reference music mixture as a query. The proposed model consists of the Single-Instrument Encoder and the Multi-Instrument Encoder, both based on convolutional neural networks. The Single-Instrument Encoder is trained to classify the instruments used in single-track audio, and we take its penultimate layer's activation as the instrument embedding. The Multi-Instrument Encoder is trained to estimate multiple instrument embeddings using the instrument embeddings computed by the Single-Instrument Encoder as a set of target embeddings. For more generalized training and realistic evaluation, we also propose a new dataset called Nlakh. Experimental results showed that the Single-Instrument Encoder was able to learn the mapping from the audio signal of unseen instruments to the instrument embedding space and the Multi-Instrument Encoder was able to extract multiple embeddings from the mixture of music and retrieve the desired instruments successfully. The code used for the experiment and audio samples are available at: https://github.com/minju0821/musical_instrument_retrieval

Sungho Lee, Matteo Scerbo, Seungu Han et al.

Geometric acoustics is an efficient framework for room acoustics modeling, governed by the canonical time-dependent rendering equation. Acoustic radiance transfer (ART) solves the equation by discretization, modeling time- and direction-dependent energy exchange between surface patches with flexible material properties. We introduce DART, an efficient, differentiable implementation of ART that enables gradient-based optimization of material properties. We evaluate DART on a simpler variant of acoustic field learning that aims to predict energy responses for novel source-receiver configurations. Experimental results demonstrate that DART generalizes better under sparse measurement scenarios than existing signal processing and neural network baselines, while maintaining simplicity and full interpretability. We open-source our implementation.

Minju Park, Kyogu Lee

Advanced music recommendation systems are being introduced along with the development of machine learning. However, it is essential to design a music recommendation system that can increase user satisfaction by understanding users' music tastes, not by the complexity of models. Although several studies related to music recommendation systems exploiting negative preferences have shown performance improvements, there was a lack of explanation on how they led to better recommendations. In this work, we analyze the role of negative preference in users' music tastes by comparing music recommendation models with contrastive learning exploiting preference (CLEP) but with three different training strategies - exploiting preferences of both positive and negative (CLEP-PN), positive only (CLEP-P), and negative only (CLEP-N). We evaluate the effectiveness of the negative preference by validating each system with a small amount of personalized data obtained via survey and further illuminate the possibility of exploiting negative preference in music recommendations. Our experimental results show that CLEP-N outperforms the other two in accuracy and false positive rate. Furthermore, the proposed training strategies produced a consistent tendency regardless of different types of front-end musical feature extractors, proving the stability of the proposed method.

Junghyun Koo, Marco A. Martínez-Ramírez, Wei-Hsiang Liao et al.

We propose an end-to-end music mixing style transfer system that converts the mixing style of an input multitrack to that of a reference song. This is achieved with an encoder pre-trained with a contrastive objective to extract only audio effects related information from a reference music recording. All our models are trained in a self-supervised manner from an already-processed wet multitrack dataset with an effective data preprocessing method that alleviates the data scarcity of obtaining unprocessed dry data. We analyze the proposed encoder for the disentanglement capability of audio effects and also validate its performance for mixing style transfer through both objective and subjective evaluations. From the results, we show the proposed system not only converts the mixing style of multitrack audio close to a reference but is also robust with mixture-wise style transfer upon using a music source separation model.

Seungyeon Rhyu, Kichang Yang, Sungjun Cho et al.

Music generation introduces challenging complexities to large language models. Symbolic structures of music often include vertical harmonization as well as horizontal counterpoint, urging various adaptations and enhancements for large-scale Transformers. However, existing works share three major drawbacks: 1) their tokenization requires domain-specific annotations, such as bars and beats, that are typically missing in raw MIDI data; 2) the pure impact of enhancing token embedding methods is hardly examined without domain-specific annotations; and 3) existing works to overcome the aforementioned drawbacks, such as MuseNet, lack reproducibility. To tackle such limitations, we develop a MIDI-based music generation framework inspired by MuseNet, empirically studying two structural embeddings that do not rely on domain-specific annotations. We provide various metrics and insights that can guide suitable encoding to deploy. We also verify that multiple embedding configurations can selectively boost certain musical aspects. By providing open-source implementations via HuggingFace, our findings shed light on leveraging large language models toward practical and reproducible music generation.

Eungbeom Kim, Jinhee Kim, Yoori Oh et al.

In this paper, we aim to unveil the impact of data augmentation in audio-language multi-modal learning, which has not been explored despite its importance. We explore various augmentation methods at not only train-time but also test-time and find out that proper data augmentation can lead to substantial improvements. Specifically, applying our proposed audio-language paired augmentation PairMix, which is the first multi-modal audio-language augmentation method, outperforms the baselines for both automated audio captioning and audio-text retrieval tasks. To fully take advantage of data augmentation, we also present multi-level test-time augmentation (Multi-TTA) for the test-time. We successfully incorporate the two proposed methods and uni-modal augmentations and achieve 47.5 SPIDEr on audio captioning, which is an 18.2% relative increase over the baseline. In audio-text retrieval, the proposed methods also show an improvement in performance as well.

Jin Woo Lee, Kyogu Lee

We present a neural network for rendering binaural speech from given monaural audio, position, and orientation of the source. Most of the previous works have focused on synthesizing binaural speeches by conditioning the positions and orientations in the feature space of convolutional neural networks. These synthesis approaches are powerful in estimating the target binaural speeches even for in-the-wild data but are difficult to generalize for rendering the audio from out-of-distribution domains. To alleviate this, we propose Neural Fourier Shift (NFS), a novel network architecture that enables binaural speech rendering in the Fourier space. Specifically, utilizing a geometric time delay based on the distance between the source and the receiver, NFS is trained to predict the delays and scales of various early reflections. NFS is efficient in both memory and computational cost, is interpretable, and operates independently of the source domain by its design. Experimental results show that NFS performs comparable to the previous studies on the benchmark dataset, even with its 25 times lighter memory and 6 times fewer calculations.

Jongho Choi, Kyogu Lee

Piano covers of pop music are enjoyed by many people. However, the task of automatically generating piano covers of pop music is still understudied. This is partly due to the lack of synchronized {Pop, Piano Cover} data pairs, which made it challenging to apply the latest data-intensive deep learning-based methods. To leverage the power of the data-driven approach, we make a large amount of paired and synchronized {Pop, Piano Cover} data using an automated pipeline. In this paper, we present Pop2Piano, a Transformer network that generates piano covers given waveforms of pop music. To the best of our knowledge, this is the first model to generate a piano cover directly from pop audio without using melody and chord extraction modules. We show that Pop2Piano, trained with our dataset, is capable of producing plausible piano covers.

Yoori Oh, Juheon Lee, Yoseob Han et al.

Recent text-to-speech models have reached the level of generating natural speech similar to what humans say. But there still have limitations in terms of expressiveness. The existing emotional speech synthesis models have shown controllability using interpolated features with scaling parameters in emotional latent space. However, the emotional latent space generated from the existing models is difficult to control the continuous emotional intensity because of the entanglement of features like emotions, speakers, etc. In this paper, we propose a novel method to control the continuous intensity of emotions using semi-supervised learning. The model learns emotions of intermediate intensity using pseudo-labels generated from phoneme-level sequences of speech information. An embedding space built from the proposed model satisfies the uniform grid geometry with an emotional basis. The experimental results showed that the proposed method was superior in controllability and naturalness.

Yunkee Chae, Junghyun Koo, Sungho Lee et al.

With the proliferation of video platforms on the internet, recording musical performances by mobile devices has become commonplace. However, these recordings often suffer from degradation such as noise and reverberation, which negatively impact the listening experience. Consequently, the necessity for music audio enhancement (referred to as music enhancement from this point onward), involving the transformation of degraded audio recordings into pristine high-quality music, has surged to augment the auditory experience. To address this issue, we propose a music enhancement system based on the Conformer architecture that has demonstrated outstanding performance in speech enhancement tasks. Our approach explores the attention mechanisms of the Conformer and examines their performance to discover the best approach for the music enhancement task. Our experimental results show that our proposed model achieves state-of-the-art performance on single-stem music enhancement. Furthermore, our system can perform general music enhancement with multi-track mixtures, which has not been examined in previous work.

Junghyun Koo, Yunkee Chae, Chang-Bin Jeon et al.

Music source separation (MSS) faces challenges due to the limited availability of correctly-labeled individual instrument tracks. With the push to acquire larger datasets to improve MSS performance, the inevitability of encountering mislabeled individual instrument tracks becomes a significant challenge to address. This paper introduces an automated technique for refining the labels in a partially mislabeled dataset. Our proposed self-refining technique, employed with a noisy-labeled dataset, results in only a 1% accuracy degradation in multi-label instrument recognition compared to a classifier trained on a clean-labeled dataset. The study demonstrates the importance of refining noisy-labeled data in MSS model training and shows that utilizing the refined dataset leads to comparable results derived from a clean-labeled dataset. Notably, upon only access to a noisy dataset, MSS models trained on a self-refined dataset even outperform those trained on a dataset refined with a classifier trained on clean labels.

Jaejun Lee, Yoori Oh, Injune Hwang et al.

This paper delves into the emerging field of face-based voice conversion, leveraging the unique relationship between an individual's facial features and their vocal characteristics. We present a novel face-based voice conversion framework that particularly utilizes the average fundamental frequency of the target speaker, derived solely from their facial images. Through extensive analysis, our framework demonstrates superior speech generation quality and the ability to align facial features with voice characteristics, including tracking of the target speaker's fundamental frequency.

Jayeon Yi, Sungho Lee, Kyogu Lee

In the heart of "rhythm games" - games where players must perform actions in sync with a piece of music - are "charts", the directives to be given to players. We newly formulate chart generation as a sequence generation task and train a Transformer using a large dataset. We also introduce tempo-informed preprocessing and training procedures, some of which are suggested to be integral for a successful training. Our model is found to outperform the baselines on a large dataset, and is also found to benefit from pretraining and finetuning.

Kyungsu Kim, Junghyun Koo, Sungho Lee et al.

Recent advancements in neural audio codecs have enabled the use of tokenized audio representations in various audio generation tasks, such as text-to-speech, text-to-audio, and text-to-music generation. Leveraging this approach, we propose TokenSynth, a novel neural synthesizer that utilizes a decoder-only transformer to generate desired audio tokens from MIDI tokens and CLAP (Contrastive Language-Audio Pretraining) embedding, which has timbre-related information. Our model is capable of performing instrument cloning, text-to-instrument synthesis, and text-guided timbre manipulation without any fine-tuning. This flexibility enables diverse sound design and intuitive timbre control. We evaluated the quality of the synthesized audio, the timbral similarity between synthesized and target audio/text, and synthesis accuracy (i.e., how accurately it follows the input MIDI) using objective measures. TokenSynth demonstrates the potential of leveraging advanced neural audio codecs and transformers to create powerful and versatile neural synthesizers. The source code, model weights, and audio demos are available at: https://github.com/KyungsuKim42/tokensynth

Jin Woo Lee, Jaehyun Park, Min Jun Choi et al.

While significant advancements have been made in music generation and differentiable sound synthesis within machine learning and computer audition, the simulation of instrument vibration guided by physical laws has been underexplored. To address this gap, we introduce a novel model for simulating the spatio-temporal motion of nonlinear strings, integrating modal synthesis and spectral modeling within a neural network framework. Our model leverages physical properties and fundamental frequencies as inputs, outputting string states across time and space that solve the partial differential equation characterizing the nonlinear string. Empirical evaluations demonstrate that the proposed architecture achieves superior accuracy in string motion simulation compared to existing baseline architectures. The code and demo are available online.

Yunkee Chae, Eunsik Shin, Suntae Hwang et al.

Lyrics generation presents unique challenges, particularly in achieving precise syllable control while adhering to song form structures such as verses and choruses. Conventional line-by-line approaches often lead to unnatural phrasing, underscoring the need for more granular syllable management. We propose a framework for lyrics generation that enables multi-level syllable control at the word, phrase, line, and paragraph levels, aware of song form. Our approach generates complete lyrics conditioned on input text and song form, ensuring alignment with specified syllable constraints. Generated lyrics samples are available at: https://tinyurl.com/lyrics9999

Yoori Oh, Yoseob Han, Kyogu Lee

There has been growing interest in audio-language retrieval research, where the objective is to establish the correlation between audio and text modalities. However, most audio-text paired datasets often lack rich expression of the text data compared to the audio samples. One of the significant challenges facing audio-text datasets is the presence of similar or identical captions despite different audio samples. Therefore, under many-to-one mapping conditions, audio-text datasets lead to poor performance of retrieval tasks. In this paper, we propose a novel approach to tackle the data imbalance problem in audio-language retrieval task. To overcome the limitation, we introduce a method that employs a distance sampling-based paraphraser leveraging ChatGPT, utilizing distance function to generate a controllable distribution of manipulated text data. For a set of sentences with the same context, the distance is used to calculate a degree of manipulation for any two sentences, and ChatGPT's few-shot prompting is performed using a text cluster with a similar distance defined by the Jaccard similarity. Therefore, ChatGPT, when applied to few-shot prompting with text clusters, can adjust the diversity of the manipulated text based on the distance. The proposed approach is shown to significantly enhance performance in audio-text retrieval, outperforming conventional text augmentation techniques.

Seonghyeon Go, Kyogu Lee

In this work, we propose a symbolic music generation model with the song structure graph analysis network. We construct a graph that uses information such as note sequence and instrument as node features, while the correlation between note sequences acts as the edge feature. We trained a Graph Neural Network to obtain node representation in the graph, then we use node representation as input of Unet to generate CONLON pianoroll image latent. The outcomes of our experimental results show that the proposed model can generate a comprehensive form of music. Our approach represents a promising and innovative method for symbolic music generation and holds potential applications in various fields in Music Information Retreival, including music composition, music classification, and music inpainting systems.

Injune Hwang, Kyogu Lee

Recently, there have been efforts to encode the linguistic information of speech using a self-supervised framework for speech synthesis. However, predicting representations from surrounding representations can inadvertently entangle speaker information in the speech representation. This paper aims to remove speaker information by exploiting the structured nature of speech, composed of discrete units like phonemes with clear boundaries. A neural network predicts these boundaries, enabling variable-length pooling for event-based representation extraction instead of fixed-rate methods. The boundary predictor outputs a probability for the boundary between 0 and 1, making pooling soft. The model is trained to minimize the difference with the pooled representation of the data augmented by time-stretch and pitch-shift. To confirm that the learned representation includes contents information but is independent of speaker information, the model was evaluated with libri-light's phonetic ABX task and SUPERB's speaker identification task.

Haesun Joung, Kyogu Lee

Music auto-tagging is crucial for enhancing music discovery and recommendation. Existing models in Music Information Retrieval (MIR) struggle with real-world noise such as environmental and speech sounds in multimedia content. This study proposes a method inspired by speech-related tasks to enhance music auto-tagging performance in noisy settings. The approach integrates Domain Adversarial Training (DAT) into the music domain, enabling robust music representations that withstand noise. Unlike previous research, this approach involves an additional pretraining phase for the domain classifier, to avoid performance degradation in the subsequent phase. Adding various synthesized noisy music data improves the model's generalization across different noise levels. The proposed architecture demonstrates enhanced performance in music auto-tagging by effectively utilizing unlabeled noisy music data. Additional experiments with supplementary unlabeled data further improves the model's performance, underscoring its robust generalization capabilities and broad applicability.

Yunkee Chae, Kyogu Lee

We present MGE-LDM, a unified latent diffusion framework for simultaneous music generation, source imputation, and query-driven source separation. Unlike prior approaches constrained to fixed instrument classes, MGE-LDM learns a joint distribution over full mixtures, submixtures, and individual stems within a single compact latent diffusion model. At inference, MGE-LDM enables (1) complete mixture generation, (2) partial generation (i.e., source imputation), and (3) text-conditioned extraction of arbitrary sources. By formulating both separation and imputation as conditional inpainting tasks in the latent space, our approach supports flexible, class-agnostic manipulation of arbitrary instrument sources. Notably, MGE-LDM can be trained jointly across heterogeneous multi-track datasets (e.g., Slakh2100, MUSDB18, MoisesDB) without relying on predefined instrument categories. Audio samples are available at our project page: https://yoongi43.github.io/MGELDM_Samples/.

Eungbeom Kim, Hantae Kim, Kyogu Lee

Transformer encoder with connectionist temporal classification (CTC) framework is widely used for automatic speech recognition (ASR). However, knowledge distillation (KD) for ASR displays a problem of disagreement between teacher-student models in frame-level alignment which ultimately hinders it from improving the student model's performance. In order to resolve this problem, this paper introduces a self-knowledge distillation (SKD) method that guides the frame-level alignment during the training time. In contrast to the conventional method using separate teacher and student models, this study introduces a simple and effective method sharing encoder layers and applying the sub-model as the student model. Overall, our approach is effective in improving both the resource efficiency as well as performance. We also conducted an experimental analysis of the spike timings to illustrate that the proposed method improves performance by reducing the alignment disagreement.

Dohoon Lee, Jaehyun Park, Hyunwoo J. Kim et al.

Flow and diffusion models have demonstrated strong performance and training stability across various tasks but lack two critical properties of simulation-based methods: freedom of dimensionality and adaptability to different inference trajectories. To address this limitation, we propose the Multidimensional Adaptive Coefficient (MAC), a plug-in module for flow and diffusion models that extends conventional unidimensional coefficients to multidimensional ones and enables inference trajectory-wise adaptation. MAC is trained via simulation-based feedback through adversarial refinement. Empirical results across diverse frameworks and datasets demonstrate that MAC enhances generative quality with high training efficiency. Consequently, our work offers a new perspective on inference trajectory optimality, encouraging future research to move beyond vector field design and to leverage training-efficient, simulation-based optimization.

Jaeyeon Kim, Injune Hwang, Kyogu Lee

We propose a framework to learn semantics from raw audio signals using two types of representations, encoding contextual and phonetic information respectively. Specifically, we introduce a speech-to-unit processing pipeline that captures two types of representations with different time resolutions. For the language model, we adopt a dual-channel architecture to incorporate both types of representation. We also present new training objectives, masked context reconstruction and masked context prediction, that push models to learn semantics effectively. Experiments on the sSIMI metric of Zero Resource Speech Benchmark 2021 and Fluent Speech Command dataset show our framework learns semantics better than models trained with only one type of representation.

Eungbeom Kim, Yunkee Chae, Jaeheon Sim et al.

Automatic speech recognition systems based on deep learning are mainly trained under empirical risk minimization (ERM). Since ERM utilizes the averaged performance on the data samples regardless of a group such as healthy or dysarthric speakers, ASR systems are unaware of the performance disparities across the groups. This results in biased ASR systems whose performance differences among groups are severe. In this study, we aim to improve the ASR system in terms of group robustness for dysarthric speakers. To achieve our goal, we present a novel approach, sample reweighting with sample affinity test (Re-SAT). Re-SAT systematically measures the debiasing helpfulness of the given data sample and then mitigates the bias by debiasing helpfulness-based sample reweighting. Experimental results demonstrate that Re-SAT contributes to improved ASR performance on dysarthric speech without performance degradation on healthy speech.

Junghyun Koo, Seungryeol Paik, Kyogu Lee

Mastering is an essential step in music production, but it is also a challenging task that has to go through the hands of experienced audio engineers, where they adjust tone, space, and volume of a song. Remastering follows the same technical process, in which the context lies in mastering a song for the times. As these tasks have high entry barriers, we aim to lower the barriers by proposing an end-to-end music remastering system that transforms the mastering style of input audio to that of the target. The system is trained in a self-supervised manner, in which released pop songs were used for training. We also anticipated the model to generate realistic audio reflecting the reference's mastering style by applying a pre-trained encoder and a projection discriminator. We validate our results with quantitative metrics and a subjective listening test and show that the model generated samples of mastering style similar to the target.

Hyeong-Seok Choi, Juheon Lee, Wansoo Kim et al.

We present a neural analysis and synthesis (NANSY) framework that can manipulate voice, pitch, and speed of an arbitrary speech signal. Most of the previous works have focused on using information bottleneck to disentangle analysis features for controllable synthesis, which usually results in poor reconstruction quality. We address this issue by proposing a novel training strategy based on information perturbation. The idea is to perturb information in the original input signal (e.g., formant, pitch, and frequency response), thereby letting synthesis networks selectively take essential attributes to reconstruct the input signal. Because NANSY does not need any bottleneck structures, it enjoys both high reconstruction quality and controllability. Furthermore, NANSY does not require any labels associated with speech data such as text and speaker information, but rather uses a new set of analysis features, i.e., wav2vec feature and newly proposed pitch feature, Yingram, which allows for fully self-supervised training. Taking advantage of fully self-supervised training, NANSY can be easily extended to a multilingual setting by simply training it with a multilingual dataset. The experiments show that NANSY can achieve significant improvement in performance in several applications such as zero-shot voice conversion, pitch shift, and time-scale modification.

Donmoon Lee, Kyogu Lee

In this study, we proposed a novel semi-supervised training method that uses unlabeled data with a class distribution that is completely different from the target data or data without a target label. To this end, we introduce a contrastive regularization that is designed to be target task-oriented and trained simultaneously. In addition, we propose an audio mixing based simple augmentation strategy that performed in batch samples. Experimental results validate that the proposed method successfully contributed to the performance improvement, and particularly showed that it has advantages in stable training and generalization.

Sungho Lee, Hyeong-Seok Choi, Kyogu Lee

Artificial reverberation (AR) models play a central role in various audio applications. Therefore, estimating the AR model parameters (ARPs) of a reference reverberation is a crucial task. Although a few recent deep-learning-based approaches have shown promising performance, their non-end-to-end training scheme prevents them from fully exploiting the potential of deep neural networks. This motivates the introduction of differentiable artificial reverberation (DAR) models, allowing loss gradients to be back-propagated end-to-end. However, implementing the AR models with their difference equations "as is" in the deep learning framework severely bottlenecks the training speed when executed with a parallel processor like GPU due to their infinite impulse response (IIR) components. We tackle this problem by replacing the IIR filters with finite impulse response (FIR) approximations with the frequency-sampling method. Using this technique, we implement three DAR models -- differentiable Filtered Velvet Noise (FVN), Advanced Filtered Velvet Noise (AFVN), and Delay Network (DN). For each AR model, we train its ARP estimation networks for analysis-synthesis (RIR-to-ARP) and blind estimation (reverberant-speech-to-ARP) task in an end-to-end manner with its DAR model counterpart. Experiment results show that the proposed method achieves consistent performance improvement over the non-end-to-end approaches in both objective metrics and subjective listening test results.

Jaejun Lee, Donmoon Lee, Hyeong-Seok Choi et al.

Ensuring performance robustness for a variety of situations that can occur in real-world environments is one of the challenging tasks in sound event classification. One of the unpredictable and detrimental factors in performance, especially in indoor environments, is reverberation. To alleviate this problem, we propose a conditioning method that provides room impulse response (RIR) information to help the network become less sensitive to environmental information and focus on classifying the desired sound. Experimental results show that the proposed method successfully reduced performance degradation caused by the reverberation of the room. In particular, our proposed method works even with similar RIR that can be inferred from the room type rather than the exact one, which has the advantage of potentially being used in real-world applications.

Junghyun Koo, Seungryeol Paik, Kyogu Lee

Reverb plays a critical role in music production, where it provides listeners with spatial realization, timbre, and texture of the music. Yet, it is challenging to reproduce the musical reverb of a reference music track even by skilled engineers. In response, we propose an end-to-end system capable of switching the musical reverb factor of two different mixed vocal tracks. This method enables us to apply the reverb of the reference track to the source track to which the effect is desired. Further, our model can perform de-reverberation when the reference track is used as a dry vocal source. The proposed model is trained in combination with an adversarial objective, which makes it possible to handle high-resolution audio samples. The perceptual evaluation confirmed that the proposed model can convert the reverb factor with the preferred rate of 64.8%. To the best of our knowledge, this is the first attempt to apply deep neural networks to converting music reverb of vocal tracks.

Hyeong-Seok Choi, Sungjin Park, Jie Hwan Lee et al.

Modern deep learning-based models have seen outstanding performance improvement with speech enhancement tasks. The number of parameters of state-of-the-art models, however, is often too large to be deployed on devices for real-world applications. To this end, we propose Tiny Recurrent U-Net (TRU-Net), a lightweight online inference model that matches the performance of current state-of-the-art models. The size of the quantized version of TRU-Net is 362 kilobytes, which is small enough to be deployed on edge devices. In addition, we combine the small-sized model with a new masking method called phase-aware $β$-sigmoid mask, which enables simultaneous denoising and dereverberation. Results of both objective and subjective evaluations have shown that our model can achieve competitive performance with the current state-of-the-art models on benchmark datasets using fewer parameters by orders of magnitude.

Sungkyun Chang, Donmoon Lee, Jeongsoo Park et al.

Most of existing audio fingerprinting systems have limitations to be used for high-specific audio retrieval at scale. In this work, we generate a low-dimensional representation from a short unit segment of audio, and couple this fingerprint with a fast maximum inner-product search. To this end, we present a contrastive learning framework that derives from the segment-level search objective. Each update in training uses a batch consisting of a set of pseudo labels, randomly selected original samples, and their augmented replicas. These replicas can simulate the degrading effects on original audio signals by applying small time offsets and various types of distortions, such as background noise and room/microphone impulse responses. In the segment-level search task, where the conventional audio fingerprinting systems used to fail, our system using 10x smaller storage has shown promising results. Our code and dataset are available at \url{https://mimbres.github.io/neural-audio-fp/}.

Junghyun Koo, Jie Hwan Lee, Jaewoo Pyo et al.

Collecting and accessing a large amount of medical data is very time-consuming and laborious, not only because it is difficult to find specific patients but also because it is required to resolve the confidentiality of a patient's medical records. On the other hand, there are deep learning models, trained on easily collectible, large scale datasets such as Youtube or Wikipedia, offering useful representations. It could therefore be very advantageous to utilize the features from these pre-trained networks for handling a small amount of data at hand. In this work, we exploit various multi-modal features extracted from pre-trained networks to recognize Alzheimer's Dementia using a neural network, with a small dataset provided by the ADReSS Challenge at INTERSPEECH 2020. The challenge regards to discern patients suspicious of Alzheimer's Dementia by providing acoustic and textual data. With the multi-modal features, we modify a Convolutional Recurrent Neural Network based structure to perform classification and regression tasks simultaneously and is capable of computing conversations with variable lengths. Our test results surpass baseline's accuracy by 18.75%, and our validation result for the regression task shows the possibility of classifying 4 classes of cognitive impairment with an accuracy of 78.70%.

Chang-Bin Jeon, Hyeong-Seok Choi, Kyogu Lee

In this paper, we propose a method of utilizing aligned lyrics as additional information to improve the performance of singing voice separation. We have combined the highway network-based lyrics encoder into Open-unmix separation network and show that the model trained with the aligned lyrics indeed results in a better performance than the model that was not informed. The question now remains whether the increase of performance is actually due to the phonetic contents that lie in the informed aligned lyrics or not. To this end, we investigated the source of performance increase in multifaceted ways by observing the change of performance when incorrect lyrics were given to the model. Experiment results show that the model can use not only just vocal activity information but also the phonetic contents from the aligned lyrics.

Hyeong-Seok Choi, Hoon Heo, Jie Hwan Lee et al.

In this work, we tackle a denoising and dereverberation problem with a single-stage framework. Although denoising and dereverberation may be considered two separate challenging tasks, and thus, two modules are typically required for each task, we show that a single deep network can be shared to solve the two problems. To this end, we propose a new masking method called phase-aware beta-sigmoid mask (PHM), which reuses the estimated magnitude values to estimate the clean phase by respecting the triangle inequality in the complex domain between three signal components such as mixture, source and the rest. Two PHMs are used to deal with direct and reverberant source, which allows controlling the proportion of reverberation in the enhanced speech at inference time. In addition, to improve the speech enhancement performance, we propose a new time-domain loss function and show a reasonable performance gain compared to MSE loss in the complex domain. Finally, to achieve a real-time inference, an optimization strategy for U-Net is proposed which significantly reduces the computational overhead up to 88.9% compared to the naïve version.

Hyeong-Seok Choi, Changdae Park, Kyogu Lee

This work seeks the possibility of generating the human face from voice solely based on the audio-visual data without any human-labeled annotations. To this end, we propose a multi-modal learning framework that links the inference stage and generation stage. First, the inference networks are trained to match the speaker identity between the two different modalities. Then the trained inference networks cooperate with the generation network by giving conditional information about the voice. The proposed method exploits the recent development of GANs techniques and generates the human face directly from the speech waveform making our system fully end-to-end. We analyze the extent to which the network can naturally disentangle two latent factors that contribute to the generation of a face image - one that comes directly from a speech signal and the other that is not related to it - and explore whether the network can learn to generate natural human face image distribution by modeling these factors. Experimental results show that the proposed network can not only match the relationship between the human face and speech, but can also generate the high-quality human face sample conditioned on its speech. Finally, the correlation between the generated face and the corresponding speech is quantitatively measured to analyze the relationship between the two modalities.

Juheon Lee, Hyeong-Seok Choi, Junghyun Koo et al.

In this study, we define the identity of the singer with two independent concepts - timbre and singing style - and propose a multi-singer singing synthesis system that can model them separately. To this end, we extend our single-singer model into a multi-singer model in the following ways: first, we design a singer identity encoder that can adequately reflect the identity of a singer. Second, we use encoded singer identity to condition the two independent decoders that model timbre and singing style, respectively. Through a user study with the listening tests, we experimentally verify that the proposed framework is capable of generating a natural singing voice of high quality while independently controlling the timbre and singing style. Also, by using the method of changing singing styles while fixing the timbre, we suggest that our proposed network can produce a more expressive singing voice.

Jie Hwan Lee, Hyeong-Seok Choi, Kyogu Lee

In recent years, music source separation has been one of the most intensively studied research areas in music information retrieval. Improvements in deep learning lead to a big progress in music source separation performance. However, most of the previous studies are restricted to separating a few limited number of sources, such as vocals, drums, bass, and other. In this study, we propose a network for audio query-based music source separation that can explicitly encode the source information from a query signal regardless of the number and/or kind of target signals. The proposed method consists of a Query-net and a Separator: given a query and a mixture, the Query-net encodes the query into the latent space, and the Separator estimates masks conditioned by the latent vector, which is then applied to the mixture for separation. The Separator can also generate masks using the latent vector from the training samples, allowing separation in the absence of a query. We evaluate our method on the MUSDB18 dataset, and experimental results show that the proposed method can separate multiple sources with a single network. In addition, through further investigation of the latent space we demonstrate that our method can generate continuous outputs via latent vector interpolation.

Juheon Lee, Hyeong-Seok Choi, Chang-Bin Jeon et al.

In this paper, we propose an end-to-end Korean singing voice synthesis system from lyrics and a symbolic melody using the following three novel approaches: 1) phonetic enhancement masking, 2) local conditioning of text and pitch to the super-resolution network, and 3) conditional adversarial training. The proposed system consists of two main modules; a mel-synthesis network that generates a mel-spectrogram from the given input information, and a super-resolution network that upsamples the generated mel-spectrogram into a linear-spectrogram. In the mel-synthesis network, phonetic enhancement masking is applied to generate implicit formant masks solely from the input text, which enables a more accurate phonetic control of singing voice. In addition, we show that two other proposed methods -- local conditioning of text and pitch, and conditional adversarial training -- are crucial for a realistic generation of the human singing voice in the super-resolution process. Finally, both quantitative and qualitative evaluations are conducted, confirming the validity of all proposed methods.

Hyeong-Seok Choi, Jang-Hyun Kim, Jaesung Huh et al.

Most deep learning-based models for speech enhancement have mainly focused on estimating the magnitude of spectrogram while reusing the phase from noisy speech for reconstruction. This is due to the difficulty of estimating the phase of clean speech. To improve speech enhancement performance, we tackle the phase estimation problem in three ways. First, we propose Deep Complex U-Net, an advanced U-Net structured model incorporating well-defined complex-valued building blocks to deal with complex-valued spectrograms. Second, we propose a polar coordinate-wise complex-valued masking method to reflect the distribution of complex ideal ratio masks. Third, we define a novel loss function, weighted source-to-distortion ratio (wSDR) loss, which is designed to directly correlate with a quantitative evaluation measure. Our model was evaluated on a mixture of the Voice Bank corpus and DEMAND database, which has been widely used by many deep learning models for speech enhancement. Ablation experiments were conducted on the mixed dataset showing that all three proposed approaches are empirically valid. Experimental results show that the proposed method achieves state-of-the-art performance in all metrics, outperforming previous approaches by a large margin.

Donmoon Lee, Jaejun Lee, Jeongsoo Park et al.

In this paper, we propose a novel method that exploits music listening log data for general-purpose music feature extraction. Despite the wealth of information available in the log data of user-item interactions, it has been mostly used for collaborative filtering to find similar items or users and was not fully investigated for content-based music applications. We resolve this problem by extending intra-domain knowledge distillation to cross-domain: i.e., by transferring knowledge obtained from the user-item domain to the music content domain. The proposed system first trains the model that estimates log information from the audio contents; then it uses the model to improve other task-specific models. The experiments on various music classification and regression tasks show that the proposed method successfully improves the performances of the task-specific models.

Sungkyun Chang, Seungjin Lee, Kyogu Lee

This paper provides an outline of the algorithms submitted for the WSDM Cup 2019 Spotify Sequential Skip Prediction Challenge (team name: mimbres). In the challenge, complete information including acoustic features and user interaction logs for the first half of a listening session is provided. Our goal is to predict whether the individual tracks in the second half of the session will be skipped or not, only given acoustic features. We proposed two different kinds of algorithms that were based on metric learning and sequence learning. The experimental results showed that the sequence learning approach performed significantly better than the metric learning approach. Moreover, we conducted additional experiments to find that significant performance gain can be achieved using complete user log information.

Juheon Lee, Seohyun Kim, Kyogu Lee

Automatic choreography generation is a challenging task because it often requires an understanding of two abstract concepts - music and dance - which are realized in the two different modalities, namely audio and video, respectively. In this paper, we propose a music-driven choreography generation system using an auto-regressive encoder-decoder network. To this end, we first collect a set of multimedia clips that include both music and corresponding dance motion. We then extract the joint coordinates of the dancer from video and the mel-spectrogram of music from audio, and train our network using music-choreography pairs as input. Finally, a novel dance motion is generated at the inference time when only music is given as an input. We performed a user study for a qualitative evaluation of the proposed method, and the results show that the proposed model is able to generate musically meaningful and natural dance movements given an unheard song.

Seungjin Lee, Juheon Lee, Kyogu lee

Interpretation of retrieved results is an important issue in music recommender systems, particularly from a user perspective. In this study, we investigate the methods for providing interpretability of content features using self-attention. We extract lyric features with the self-attentive genre classification model trained on 140,000 tracks of lyrics. Likewise, we extract acoustic features using the acoustic model with self-attention trained on 120,000 tracks of acoustic signals. The experimental results show that the proposed methods provide the characteristics that are interpretable in terms of both lyrical and musical contents. We demonstrate this by visualizing the attention weights, and by presenting the most similar songs found using lyric or audio features.

Sungheon Park, Taehoon Kim, Kyogu Lee et al.

In this paper, we propose a simple yet effective method for multiple music source separation using convolutional neural networks. Stacked hourglass network, which was originally designed for human pose estimation in natural images, is applied to a music source separation task. The network learns features from a spectrogram image across multiple scales and generates masks for each music source. The estimated mask is refined as it passes over stacked hourglass modules. The proposed framework is able to separate multiple music sources using a single network. Experimental results on MIR-1K and DSD100 datasets validate that the proposed method achieves competitive results comparable to the state-of-the-art methods in multiple music source separation and singing voice separation tasks.

Jeongsoo Park, Jaeyoung Shin, Kyogu Lee

Spectral envelope is one of the most important features that characterize the timbre of an instrument sound. However, it is difficult to use spectral information in the framework of conventional spectrogram decomposition methods. We overcome this problem by suggesting a simple way to provide a constraint on the spectral envelope calculated by linear prediction. In the first part of this study, we use a pre-trained spectral envelope of known instruments as the constraint. Then we apply the same idea to a blind scenario in which the instruments are unknown. The experimental results reveal that the proposed method outperforms the conventional methods.

Hyungui Lim, Seungyeon Rhyu, Kyogu Lee

Generating a chord progression from a monophonic melody is a challenging problem because a chord progression requires a series of layered notes played simultaneously. This paper presents a novel method of generating chord sequences from a symbolic melody using bidirectional long short-term memory (BLSTM) networks trained on a lead sheet database. To this end, a group of feature vectors composed of 12 semitones is extracted from the notes in each bar of monophonic melodies. In order to ensure that the data shares uniform key and duration characteristics, the key and the time signatures of the vectors are normalized. The BLSTM networks then learn from the data to incorporate the temporal dependencies to produce a chord progression. Both quantitative and qualitative evaluations are conducted by comparing the proposed method with the conventional HMM and DNN-HMM based approaches. Proposed model achieves 23.8% and 11.4% performance increase from the other models, respectively. User studies further confirm that the chord sequences generated by the proposed method are preferred by listeners.

Sungkyun Chang, Juheon Lee, Sang Keun Choe et al.

In this paper, we propose a new approach to cover song identification using a CNN (convolutional neural network). Most previous studies extract the feature vectors that characterize the cover song relation from a pair of songs and used it to compute the (dis)similarity between the two songs. Based on the observation that there is a meaningful pattern between cover songs and that this can be learned, we have reformulated the cover song identification problem in a machine learning framework. To do this, we first build the CNN using as an input a cross-similarity matrix generated from a pair of songs. We then construct the data set composed of cover song pairs and non-cover song pairs, which are used as positive and negative training samples, respectively. The trained CNN outputs the probability of being in the cover song relation given a cross-similarity matrix generated from any two pieces of music and identifies the cover song by ranking on the probability. Experimental results show that the proposed algorithm achieves performance better than or comparable to the state-of-the-art.