Publications

Theses

2021

1. B.Sc.Music

   Bridging Transformers and latent variable models for user-controllable conditional music generation

   Wu, Shih-Lun

   Bachelor’s Thesis, National Taiwan University 2021

   PDF Slides

Peer-reviewed Journal & Conference Papers

2024

1. ISMIRMusic

   Audio Prompt Adapter: Unleashing music editing abilities for text-to-music with lightweight finetuning

   Tsai, Fang-Duo,  Wu, Shih-Lun,  Kim, Haven and 3 more authors

   In International Society for Music Information Retrieval Conference 2024

   Abstract PDF Code Website

   Text-to-music models allow users to generate nearly realistic musical audio with textual commands. However, editing music audios remains challenging due to the conflicting desiderata of performing fine-grained alterations on the audio while maintaining a simple user interface. To address this challenge, we propose Audio Prompt Adapter (or AP-Adapter), a lightweight addition to pretrained text-to-music models. We utilize AudioMAE to extract features from the input audio, and construct attention-based adapters to feedthese features into the internal layers of AudioLDM2, a diffusion-based text-to-music model. With 22M trainable parameters, AP-Adapter empowers users to harness both global (e.g., genre and timbre) and local (e.g., melody) aspects of music, using the original audio and a short text as inputs. Through objective and subjective studies, we evaluate AP-Adapter on three tasks: timbre transfer, genre transfer, and accompaniment generation. Additionally, we demonstrate its effectiveness on out-of-domain audios containing unseen instruments during training.

2. TASLPMusic

   Music ControlNet: Multiple time-varying controls for music generation

   Wu, Shih-Lun, Donahue, Chris,  Watanabe, Shinji and 1 more author

   IEEE/ACM Transactions on Audio, Speech, & Language Processing 2024

   Abstract PDF Website TL;DR Code (3rd-party)

   Text-to-music generation models are now capable of generating high-quality music audio in broad styles. However, text control is primarily suitable for the manipulation of global musical attributes like genre, mood, and tempo, and is less suitable for precise control over time-varying attributes such as the positions of beats in time or the changing dynamics of the music. We propose Music ControlNet, a diffusion-based music generation model that offers multiple precise, time-varying controls over generated audio. To imbue text-to-music models with time-varying control, we propose an approach analogous to pixel-wise control of the image-domain ControlNet method. Specifically, we extract controls from training audio yielding paired data, and fine-tune a diffusion-based conditional generative model over audio spectrograms given melody, dynamics, and rhythm controls. While the image-domain Uni-ControlNet method already allows generation with any subset of controls, we devise a new strategy to allow creators to input controls that are only partially specified in time. We evaluate both on controls extracted from audio and controls we expect creators to provide, demonstrating that we can generate realistic music that corresponds to control inputs in both settings. While few comparable music generation models exist, we benchmark against MusicGen, a recent model that accepts text and melody input, and show that our model generates music that is 49 percent more faithful to input melodies despite having 35x fewer parameters, training on 11x less data, and enabling two additional forms of time-varying control.

3. ICASSPAudio

   Improving audio captioning models with fine-grained audio features, text embedding supervision, and LLM mix-up augmentation

   Wu, Shih-Lun, Chang, Xuankai,  Wichern, Gordon and 4 more authors

   International Conference on Acoustics, Speech, & Signal Processing 2024

   Abstract PDF

   Automated audio captioning (AAC) aims to generate informative descriptions for various sounds from nature and/or human activities. In recent years, AAC has quickly attracted research interest, with state-of-the-art systems now relying on a sequence-to-sequence (seq2seq) backbone powered by strong models such as Transformers. Following the macro-trend of applied machine learning research, in this work, we strive to improve the performance of seq2seq AAC models by extensively leveraging pretrained models and large language models (LLMs). Specifically, we utilize BEATs to extract fine-grained audio features. Then, we employ Instructor LLM to fetch text embeddings of captions, and infuse their language-modality knowledge into BEATs audio features via an auxiliary InfoNCE loss function. Moreover, we propose a novel data augmentation method that uses ChatGPT to produce caption mix-ups (i.e., grammatical and compact combinations of two captions) which, together with the corresponding audio mixtures, increase not only the amount but also the complexity and diversity of training data. During inference, we propose to employ nucleus sampling and a hybrid reranking algorithm, which has not been explored in AAC research. Combining our efforts, our model achieves a new state-of-the-art 32.6 SPIDEr-FL score on the Clotho evaluation split, and wins the 2023 DCASE AAC challenge.

2023

1. ACLMultimodal

   Listener model for the PhotoBook referential game with CLIPScores as implicit reference chain

   Wu, Shih-Lun, Chou, Yi-Hui,  and Li, Liangze

   In Annual Meeting of the Association for Computational Linguistics 2023

   Abstract PDF Code

   PhotoBook is a collaborative dialogue game where two players receive private, partially-overlapping sets of images and resolve which images they have in common. It presents machines with a great challenge to learn how people build common ground around multimodal context to communicate effectively. Methods developed in the literature, however, cannot be deployed to real gameplay since they only tackle some subtasks of the game, and they require additional reference chains inputs, whose extraction process is imperfect. Therefore, we propose a reference chain-free listener model that directly addresses the game’s predictive task, i.e., deciding whether an image is shared with partner. Our DeBERTa-based listener model reads the full dialogue, and utilizes CLIPScore features to assess utterance-image relevance. We achieve >77 percent accuracy on unseen sets of images/game themes, outperforming baseline by >17 points.

2. ICASSPMusic

   Compose & Embellish: Well-structured piano performance generation via a two-stage approach

   Wu, Shih-Lun,  and Yang, Yi-Hsuan

   In International Conference on Acoustics, Speech, & Signal Processing 2023

   Abstract PDF Code

   Even with strong sequence models like Transformers, generating expressive piano performances with long-range musical structures remains challenging. Meanwhile, methods to compose well-structured melodies or lead sheets (melody + chords), i.e., simpler forms of music, gained more success. Observing the above, we devise a two-stage Transformer-based framework that Composes a lead sheet first, and then Embellishes it with accompaniment and expressive touches. Such a factorization also enables pretraining on non-piano data. Our objective and subjective experiments show that Compose & Embellish shrinks the gap in structureness between a current state of the art and real performances by half, and improves other musical aspects such as richness and coherence as well.

2022

1. TASLPMusic

   MuseMorphose: Full-song and fine-grained piano music style transfer with one Transformer VAE

   Wu, Shih-Lun,  and Yang, Yi-Hsuan

   IEEE/ACM Transactions on Audio, Speech, & Language Processing 2023

   Abstract PDF Code Website

   Transformers and variational autoencoders (VAE) have been extensively employed for symbolic (e.g., MIDI) domain music generation. While the former boast an impressive capability in modeling long sequences, the latter allow users to willingly exert control over different parts (e.g., bars) of the music to be generated. In this paper, we are interested in bringing the two together to construct a single model that exhibits both strengths. The task is split into two steps. First, we equip Transformer decoders with the ability to accept segment-level, time-varying conditions during sequence generation. Subsequently, we combine the developed and tested in-attention decoder with a Transformer encoder, and train the resulting MuseMorphose model with the VAE objective to achieve style transfer of long pop piano pieces, in which users can specify musical attributes including rhythmic intensity and polyphony (i.e., harmonic fullness) they desire, down to the bar level. Experiments show that MuseMorphose outperforms recurrent neural network (RNN) based baselines on numerous widely-used metrics for style transfer tasks.

2. TMMMusic

   Theme Transformer: Symbolic music generation with theme-conditioned Transformer

   Shih, Y.-J.,  Wu, S.-L.,  Zalkow, F. and 2 more authors

   IEEE Transactions on Multimedia 2022

   Abstract PDF Code Website

   Attention-based Transformer models have been increasingly employed for automatic music generation. To condition the generation process of such a model with a user-specified sequence, a popular approach is to take that conditioning sequence as a priming sequence and ask a Transformer decoder to generate a continuation. However, this prompt-based conditioning cannot guarantee that the conditioning sequence would develop or even simply repeat itself in the generated continuation. In this paper, we propose an alternative conditioning approach, called theme-based conditioning, that explicitly trains the Transformer to treat the conditioning sequence as a thematic material that has to manifest itself multiple times in its generation result. This is achieved with two main technical contributions. First, we propose a deep learning-based approach that uses contrastive representation learning and clustering to automatically retrieve thematic materials from music pieces in the training data. Second, we propose a novel gated parallel attention module to be used in a sequence-to-sequence (seq2seq) encoder/decoder architecture to more effectively account for a given conditioning thematic material in the generation process of the Transformer decoder. We report on objective and subjective evaluations of variants of the proposed Theme Transformer and the conventional prompt-based baseline, showing that our best model can generate, to some extent, polyphonic pop piano music with repetition and plausible variations of a given condition.

2021

1. ICMLML

   Relative positional encoding for Transformers with linear complexity

   Liutkus*, A., Cı́fka*, O.,  Wu, S.-L. and 3 more authors

   In International Conference on Machine Learning (long talk) 2021

   Abstract PDF Code Website Talk

   Recent advances in Transformer models allow for unprecedented sequence lengths, due to linear space and time complexity. In the meantime, relative positional encoding (RPE) was proposed as beneficial for classical Transformers and consists in exploiting lags instead of absolute positions for inference. Still, RPE is not available for the recent linear-variants of the Transformer, because it requires the explicit computation of the attention matrix, which is precisely what is avoided by such methods. In this paper, we bridge this gap and present Stochastic Positional Encoding as a way to generate PE that can be used as a replacement to the classical additive (sinusoidal) PE and provably behaves like RPE. The main theoretical contribution is to make a connection between positional encoding and cross-covariance structures of correlated Gaussian processes. We illustrate the performance of our approach on the Long-Range Arena benchmark and on music generation.

2020

1. ICMLAVision

   Deep learning for automatic quality grading of mangoes: Methods and insights

   Wu, Shih-Lun, Tung, Hsiao-Yen,  and Hsu, Yu-Lun

   In International Conference on Machine Learning and Applications 2020

   Abstract PDF Talk

   The quality grading of mangoes is a crucial task for mango growers as it vastly affects their profit. However, until today, this process still relies on laborious efforts of humans, who are prone to fatigue and errors. To remedy this, the paper approaches the grading task with various convolutional neural networks (CNN), a tried-and-tested deep learning technology in computer vision. The models involved include Mask R-CNN (for background removal), the numerous past winners of the ImageNet challenge, namely AlexNet, VGGs, and ResNets; and, a family of self-defined convolutional autoencoder-classifiers (ConvAE-Clfs) inspired by the claimed benefit of multi-task learning in classification tasks. Transfer learning is also adopted in this work via utilizing the ImageNet pretrained weights. Besides elaborating on the preprocessing techniques, training details, and the resulting performance, we go one step further to provide explainable insights into the model’s working with the help of saliency maps and principal component analysis (PCA). These insights provide a succinct, meaningful glimpse into the intricate deep learning black box, fostering trust, and can also be presented to humans in real-world use cases for reviewing the grading results.

2. ISMIRMusic

   The Jazz Transformer on the front line: Exploring the shortcomings of AI-composed music through quantitative measures

   Wu, Shih-Lun,  and Yang, Yi-Hsuan

   In International Society for Music Information Retrieval Conference 2020

   Abstract PDF Supp Code Talk

   This paper presents the Jazz Transformer, a generative model that utilizes a neural sequence model called the Transformer-XL for modeling lead sheets of Jazz music. Moreover, the model endeavors to incorporate structural events present in the Weimar Jazz Database (WJazzD) for inducing structures in the generated music. While we are able to reduce the training loss to a low value, our listening test suggests however a clear gap between the average ratings of the generated and real compositions. We therefore go one step further and conduct a series of computational analysis of the generated compositions from different perspectives. This includes analyzing the statistics of the pitch class, grooving, and chord progression, assessing the structureness of the music with the help of the fitness scape plot, and evaluating the model’s understanding of Jazz music through a MIREX-like continuation prediction task. Our work presents in an analytical manner why machine-generated music to date still falls short of the artwork of humanity, and sets some goals for future work on automatic composition to further pursue.

3. RVVerification

   Efficient system verification with multiple weakly-hard constraints for runtime monitoring

   Wu*, S.-L., Bai*, C.-Y.,  Chang, K.-C. and 5 more authors

   In International Conference on Runtime Verification 2020

   Abstract HTML PDF

   A weakly-hard fault model can be captured by an (m, k) constraint, where 0≤m≤k, meaning that there are at most m bad events (faults) among any k consecutive events. In this paper, we use a weakly-hard fault model to constrain the occurrences of faults in system inputs. We develop approaches to verify properties for all possible values of (m, k), where k is smaller than or equal to a given K, in an exact and efficient manner. By verifying all possible values of (m, k), we define weakly-hard requirements for the system environment and design a runtime monitor based on counting the number of faults in system inputs. If the system environment satisfies the weakly-hard requirements, the satisfaction of desired properties is guaranteed; otherwise, the runtime monitor can notify the system to switch to a safe mode. Experimental results with a discrete second-order controller demonstrate the efficiency of the proposed approaches.