Generative speech enhancement methods based on generative adversarial networks (GANs) have demonstrated promising performance across various speech enhancement tasks. However, their performance in very low signal-to-noise ratio (SNR) scenarios remains under-explored and limited, as these conditions pose significant challenges to both discriminative and generative state-of-the-art methods. To address this, we propose DisCoGAN, a GAN-based speech enhancement method that leverages latent features extracted from discriminative speech enhancement models as generic conditioning information. By incorporating the proposed discriminative conditioning method, DisCoGAN improves speech quality and intelligibility, particularly in low-SNR scenarios, while maintaining competitive or superior performance in high-SNR conditions and real-world recordings. We also conduct a comprehensive evaluation of conventional GAN-based architectures, including end-to-end GANs, GAN-first, and post-filtering GANs, as well as discriminative models under low-SNR conditions, and show that DisCoGAN consistently outperforms existing methods. Finally, we present ablation studies that highlight the performance gains from discriminative conditioning and demonstrate how DisCoGAN leverages both local and global temporal context, providing insight into the key factors underlying these gains.
The Real World Computing (RWC) Music Database has been a cornerstone of Music Information Retrieval (MIR) research for over two decades, offering high-quality record- ings across multiple genres, including popular, classical , jazz music. Beyond its extensive audio collection, the dataset is enriched by aligned Musical Instrument Digital Interface (MIDI) encodings , complementary annotations, including beat, structure, and chord labels, making it a valuable resource for music structure analysis, beat track- ing, chord recognition, automatic transcription, and music synchronization. Originally, the RWC audio material was distributed on physical media and made available for purchase at a nominal price. A significant development, announced and initiated with this paper, is the release of the RWC dataset under a Creative Commons license, mak- ing it freely accessible for research purposes. This transition significantly enhances the dataset's usability and supports broader adoption within the MIR research community. We outline the steps taken to enable this release and share a vision for transforming RWC into a community-driven resource that promotes open research and collabora- tion. With the audio recordings now hosted on Zenodo, we also discuss strategies for dataset maintenance, annotation expansion, and reproducibility through collabora- tive platforms such as GitHub. This shift promotes transparency and inclusivity, helping to ensure the dataset's continued relevance for cutting-edge MIR research. We fur- ther revisit the historical significance of the RWC dataset, incorporating insights from an interview with its original creator, Masataka Goto, and provide an overview of its current applications and future potential. In summary, by embracing an open and community-supported approach, we aim not only to renew the dataset's impact and preserve its legacy within the MIR community but also to shed light on broader best practices for open, collaborative, and sustainable research infrastructures.
This paper presents the Deep learning-based Perceptual Audio Quality metric (DeePAQ) for evaluating general audio quality. Our approach leverages metric learning together with the music foundation model MERT, guided by surrogate labels, to construct an embedding space that captures distortion intensity in general audio. To the best of our knowledge, DeePAQ is the first in the general audio quality domain to leverage weakly supervised labels and metric learning for fine-tuning a music foundation model with Low-Rank Adaptation (LoRA), a direction not yet explored by other state-of-the-art methods. We benchmark the proposed model against state-of-the-art objective audio quality metrics across listening tests spanning audio coding and source separation. Results show that our method surpasses existing metrics in detecting coding artifacts and generalizes well to unseen distortions such as source separation, highlighting its robustness and versatility.
This paper presents an approach for acoustic teleportation by disentangling speech content from acoustic environment characteristics in neural audio codec representations. Acoustic teleportation transfers room characteristics between speech recordings while preserving content and speaker identity, enabling applications in telecommunications and virtual acoustic environments. Extending prior work by adopting the EnCodec architecture, we achieve substantial objective quality improvements with non-intrusive ScoreQ scores of 3.03, compared to 2.44 for previous methods. Our training strategy incorporates five tasks: clean reconstruction, reverberated reconstruction, dereverberation, and two variants of acoustic teleportation. We analyze the trade-off between temporal downsampling of acoustic embeddings and reconstruction quality, demonstrating that even a downsampling factor of two yields a statistically significant degradation in reconstruction quality. The learned acoustic embeddings exhibit a strong correlation with reverberation time. t-SNE analysis reveals that acoustic embeddings cluster predominantly by room while speech embeddings cluster by speaker, confirming substantial disentanglement sufficient for effective acoustic teleportation.
Precise note-level annotations are critical for training automatic music transcription (AMT) systems, in particular note-onset labels, which form a core component of many recent AMT systems. However, high-quality annotations for real-world recordings are scarce. Sequence-level score–audio alignment methods such as dynamic time warping provide only coarse correspondence, making a local refinement step necessary. This refinement step, known as snapping, adjusts aligned score onsets using peaks in a neural onset posteriorgram and often determines whether weakly aligned score–audio pairs become usable training data at all. Despite its practical importance, snapping is typically treated as a simple post-processing heuristic and implemented with greedy local decisions. We present a systematic analysis of snapping strategies for training instrument-agnostic transcribers, demonstrating that snapping is essential for learning from weakly aligned data. Building on this, we formulate snapping as a per-pitch assignment problem and solve it via bipartite graph matching, yielding context-aware onset decisions under overlapping refinement windows and uncertain initial alignments. Extensive cross-dataset experiments across piano, chamber, and orchestral recordings show improved onset alignment and transcription accuracy over greedy snapping, with gains increasing for wider snapping windows and coarser initial alignments. Qualitative examples are provided on our project page: https://abhirupsaha8.github.io
