Mainstream zero-shot TTS production systems like Voicebox and Seed-TTS achieve human parity speech by leveraging Flow-matching and Diffusion models, respectively. Unfortunately, human-level audio synthesis leads to identity misuse and information security issues. Currently, many antispoofing models have been developed against deepfake audio. However, the efficacy of current state-of-the-art anti-spoofing models in countering audio synthesized by diffusion and flowmatching based TTS systems remains unknown. In this paper, we proposed the Diffusion and Flow-matching based Audio Deepfake (DFADD) dataset. The DFADD dataset collected the deepfake audio based on advanced diffusion and flowmatching TTS models. Additionally, we reveal that current anti-spoofing models lack sufficient robustness against highly human-like audio generated by diffusion and flow-matching TTS systems. The proposed DFADD dataset addresses this gap and provides a valuable resource for developing more resilient anti-spoofing models.

Recent years have seen increasing interest in applying deep learning methods to the modeling of guitar amplifiers or effect pedals. Existing methods are mainly based on the supervised approach, requiring temporally-aligned data pairs of unprocessed and rendered audio. However, this approach does not scale well, due to the complicated process involved in creating the data pairs. A very recent work done by Wright et al. has explored the potential of leveraging unpaired data for training, using a generative adversarial network (GAN)-based framework. This paper extends their work by using more advanced discriminators in the GAN, and using more unpaired data for training. Specifically, drawing inspiration from recent advancements in neural vocoders, we employ in our GAN-based model for guitar amplifier modeling two sets of discriminators, one based on multi-scale discriminator (MSD) and the other multi-period discriminator (MPD). Moreover, we experiment with adding unprocessed audio signals that do not have the corresponding rendered audio of a target tone to the training data, to see how much the GAN model benefits from the unpaired data. Our experiments show that the proposed two extensions contribute to the modeling of both low-gain and high-gain guitar amplifiers.

We introduce MIR-MLPop, a publicly available multilingual pop music dataset designed for automatic lyrics transcription and lyrics alignment in polyphonic music. The dataset comprises 90 pop music tracks in Mandarin, Cantonese, and Taiwanese Hokkien, with manually annotated time-aligned lyrics with both characters and pronunciation labels. To the best of our knowledge, this is the first ever singing dataset for Cantonese and Taiwanese Hokkien. In the experiments, using the pretrained Whisper model as the backbone, we develop lyrics transcription and lyrics alignment models for all three languages. Overall, the results are promising for both tasks, but show clear differences among the languages. Our models perform significantly better on languages that have been seen by Whisper during pretraining than on the language unseen by Whisper. This finding highlights the potential challenge in lyrics transcription and alignment for low-resource languages that have not been covered by pretrained speech models.

Audio-visual synchronization aims to determine whether the mouth movements and speech in the video are synchronized. VocaLiST reaches state-of-the-art performance by incorporating multimodal Transformers to model audio-visual interact information. However, it requires high computing resources, making it impractical for real-world applications. This paper proposed an MTDVocaLiST model, which is trained by our proposed multimodal Transformer distillation (MTD) loss. MTD loss enables MTDVocaLiST model to deeply mimic the cross-attention distribution and value-relation in the Transformer of VocaLiST. Additionally, we harness uncertainty weighting to fully exploit the interaction information across all layers. Our proposed method is effective in two aspects: From the distillation method perspective, MTD loss outperforms other strong distillation baselines. From the distilled model's performance perspective: 1) MTDVocaLiST outperforms similar-size SOTA models, SyncNet, and Perfect Match models by 15.65% and 3.35%; 2) MTDVocaLiST reduces the model size of VocaLiST by 83.52%, yet still maintaining similar performance.

Replicating analog device circuits through neural audio effect modeling has garnered increasing interest in recent years. Existing work has predominantly focused on a one-to-one emulation strategy, modeling specific devices individually. In this paper, we tackle the less-explored scenario of one-to-many emulation, utilizing conditioning mechanisms to emulate multiple guitar amplifiers through a single neural model. For condition representation, we use contrastive learning to build a tone embedding encoder that extracts style-related features of various amplifiers, leveraging a dataset of comprehensive amplifier settings. Targeting zero-shot application scenarios, we also examine various strategies for tone embedding representation, evaluating referenced tone embedding against two retrieval-based embedding methods for amplifiers unseen in the training time. Our findings showcase the efficacy and potential of the proposed methods in achieving versatile one-to-many amplifier modeling, contributing a foundational step towards zero-shot audio modeling applications.

Altered gut metabolites, especially short-chain fatty acids (SCFAs), in feces and plasma are observed in patients with Parkinson’s disease (PD). We aimed to investigate the colonic expression of two SCFA receptors, free fatty acid receptor (FFAR)2 and FFAR3, and gut barrier integrity in patients with PD and correlations with clinical severity. In this retrospective study, colonic biopsy specimens were collected from 37 PD patients and 34 unaffected controls. Of this cohort, 31 participants (14 PD, 17 controls) underwent a series of colon biopsies. Colonic expression of FFAR2, FFAR3, and the tight junction marker ZO-1 were assayed by immunofluorescence staining. The You Only Look Once (version 8, YOLOv8) algorithm was used for automated detection and segmentation of immunostaining signal. PD motor function was assessed with the Movement Disorder Society (MDS)-Unified Parkinson’s Disease Rating Scale (UPDRS), and constipation was assessed using Rome-IV criteria. Compared with controls, PD patients had significantly lower colonic expression of ZO-1 (p < 0.01) and FFAR2 (p = 0.01). On serial biopsy, colonic expression of FFAR2 and FFAR3 was reduced in the pre-motor stage before PD diagnosis (both p < 0.01). MDS-UPDRS motor scores did not correlate with colonic marker levels. Constipation severity negatively correlated with colonic ZO-1 levels (r = -0.49, p = 0.02). Colonic expression of ZO-1 and FFAR2 is lower in PD patients compared with unaffected controls, and FFAR2 and FFAR3 levels decline in the pre-motor stage of PD. Our findings implicate a leaky gut phenomenon in PD and reinforce that gut metabolites may contribute to the process of PD.

Optical side-channel analysis poses a significant threat to the security of integrated circuits (ICs) by enabling the disclosure of secret data, such as encryption keys. In this paper, for the first time, we present a multiphysics simulation framework of optical side-channel analysis from the layout database of a fabricated testchip. By leveraging accurate device models and electro-photonic physics, our framework models the photon emission behavior in ICs and enables the statistical correlation of emitted photon patterns with secret keys. Our framework enhances understanding of layout-level optical side-channel leakage and its implications, enabling IC designers to assess the risks associated with optical side-channel attacks and develop efficient countermeasures at the pre-silicon stage.

Music source separation aims to separate polyphonic music into different types of sources. Most existing methods focus on enhancing the quality of separated results by using a larger model structure, rendering them unsuitable for deployment on edge devices. Moreover, these methods may produce low-quality output when the input duration is short, making them impractical for real-time applications. This challenge is akin to those in speech processing models and systems, where isolating and analyzing specific audio components is critical. Therefore, the goal of this paper is to enhance a lightweight model, MMDenstNet, to strike a balance between separation quality and latency for real-time applications. Different directions of improvement are explored or proposed in this paper, including complex ideal ratio mask, self-attention, band-merge-split method, and feature look back. Source-to-distortion ratio, real-time factor, and optimal latency are employed to evaluate the performance. To align with our application requirements, the evaluation process in this paper focuses on the separation performance of the accompaniment part. Experimental results demonstrate that our improvements achieve a low real-time factor and optimal latency while maintaining a comparable source-to-distortion ratio.

Due to the rapid development of deep learning, we can now successfully separate singing voice from mono audio music. However, this separation can only extract human voices from other musical instruments, which is undesirable for karaoke content generation applications that only require the separation of lead singers. For this karaoke application, we need to separate the music containing male and female duets into two vocals, or extract a single lead vocal from the music containing vocal harmony. For this reason, we propose in this article to use a singer separation system, which generates karaoke content for one or two separated lead singers. In particular, we introduced three models for the singer separation task and designed an automatic model selection scheme to distinguish how many lead singers are in the song. We also collected a large enough data set, MIR-SingerSeparation (1), which has been publicly released to advance the frontier of this research. Our singer separation is most suitable for sentimental ballads and can be directly applied to karaoke content generation. As far as we know, this is the first singer-separation work for real-world karaoke applications.

We present the results of the search for an isotropic stochastic gravitational wave background (GWB) at nanohertz frequencies using the second data release of the European Pulsar Timing Array (EPTA) for 25 millisecond pulsars and a combination with the first data release of the Indian Pulsar Timing Array (InPTA). We analysed (i) the full 24.7-year EPTA data set, (ii) its 10.3-year subset based on modern observing systems, (iii) the combination of the full data set with the first data release of the InPTA for ten commonly timed millisecond pulsars, and (iv) the combination of the 10.3-year subset with the InPTA data. These combinations allowed us to probe the contributions of instrumental noise and interstellar propagation effects. With the full data set, we find marginal evidence for a GWB, with a Bayes factor of four and a false alarm probability of $4\%$. With the 10.3-year subset, we report evidence for a GWB, with a Bayes factor of $60$ and a false alarm probability of about $0.1\%$ ($\gtrsim 3\sigma$ significance). The addition of the InPTA data yields results that are broadly consistent with the EPTA-only data sets, with the benefit of better noise modelling. Analyses were performed with different data processing pipelines to test the consistency of the results from independent software packages. The inferred spectrum from the latest EPTA data from new generation observing systems is rather uncertain and in mild tension with the common signal measured in the full data set. However, if the spectral index is fixed at 13/3, the two data sets give a similar amplitude of ($2.5\pm0.7)\times10^{-15}$ at a reference frequency of $1\,{\rm yr}^{-1}$. By continuing our detection efforts as part of the International Pulsar Timing Array (IPTA), we expect to be able to improve the measurement of spatial correlations and better characterise this signal in the coming years.

Background: Nigrosome-1 imaging has been used for assisting the diagnosis of Parkinson's disease (PD). We aimed to examine the diagnostic performance of loss of nigrosome-1 in PD and the correlation between the size of the nigrosome-1 and motor severity of PD.Methods: We included 237 patients with PD and 165 controls. The motor severity of PD was assessed with the Unified Parkinson's Disease Rating Scale (UPDRS) part III score and Hoehn-Yahr staging. The 3 or 1.5 Tesla susceptibility-weighted imaging combined with a deep-learning algorithm was applied for detecting the loss and the size of nigrosome-1. Clinical correlations and diagnostic performance of size of nigrosome-1 were also investigated.Results: The mean nigrosome-1 size was significantly smaller in PD patients than in controls (0.06 & PLUSMN; 0.07 cm2 vs. 0.20 & PLUSMN; 0.05 cm2, P < 0.001). The area under the receiver operating characteristic curve (AUC) of the established model showed 0.94 accuracy (95% confidence interval [CI]: 0.87, 1.01, P < 0.01) in differentiating between the PD and control groups. Moreover, the partial loss of nigrosome-1 detected with SWI had an AUC of 0.96 in discriminating early-stage PD from controls (95% CI: 0.88, 1.02, P < 0.001). After adjusting for age, sex, disease duration, and levodopa equivalent daily dose, the estimated size of nigrosome-1 was negatively associated with the UPDRS part III motor score (& rho; =-0.433, P < 0.001), but not with Mini-Mental State Examination scores (& rho; = 0.006, P = 0.894).Conclusions: The extent of loss and the size of nigrosome-1 may potentially assist in the diagnosis of PD. Nigrosome-1 size reflects the motor severity of PD.

Predicting the next interaction based on an anony-mous short-term sequence is challenging in session-based rec-ommendation. Multi-behavior recommendations aim to capture effective user intention representations by considering session sequences with several action types. However, recent multi-behavior-based approaches for session-based recommendation still have limitations. First, the final prediction for most existing approaches is limited to the next item, ignoring which action the predicted item is associated with. Second, existing approaches consider item sequences and action sequences individually and thus do not explicitly model the action dependencies for a single item. In this paper, we propose a novel session-based recommendation algorithm with Action Pattern-Aware Networks (APANet), which could incorporate both historical item sequences and reformulated item-wise action patterns into the modeling process, and predict the next-best interaction (i.e., next-best item and its associated action) given a short-term anonymous multi-behavior sequence. Comprehensive experiments on three public benchmark datasets demonstrate the effectiveness of the proposed APANet.

Zero-shot singing voice synthesis (SVS), the task to synthesize the singing voice of an arbitrary target singer, has gained increasing attentions in the past few years. Several recently proposed systems have demonstrated promising results on this task. However, these systems require detailed musical features at the frame level as the musical content. To deal with this issue, we propose a model that performs zero-shot SVS with only musical score as the musical content condition. To help model training, we build an acoustic encoder that extracts linguistic features from audio, and train it with the lyrics transcription objective. The output of the acoustic encoder serves as an alternative to the musical score, allowing the SVS model to learn from weakly labeled data. Results suggest that the proposed method outperforms baseline semi-supervised method in both subjective and objective tests.

Photoaging mainly occurs due to ultraviolet (UV) radiation, and is accompanied by increased secretion of matrix metalloproteinases (MMPs) and degradation of collagen. UV radiation induces cell senescence in the skin; however, the role of senescent cells in photoaging remains unclear. Therefore, to elucidate the role of senescent cells in photoaging, we evaluated the effect of senolytics in a photoaging mouse model and investigated the underlying mechanism of their antiaging effect. In both UV-induced senescent human dermal fibroblasts and a photoaging mouse model, ABT-263 and ABT-737 showed senolytic effects on senescent fibroblasts. Moreover, we found that several senescence-associated secretory phenotype factors, such as IL-6, CCL5, CCL7, CXCL12, and SCF, induced MMP-1 expression in dermal fibroblasts, which decreased after treatment with ABT-263 and ABT-737in vivo and in vitro. Both senolytic drugs attenuated the induction of MMPs and the decrease of collagen density in the photoaging mouse model. Our data suggest that targeting senescent dermal fibroblasts may reduce UV-induced skin aging phenotypes and thus senolytic agents are promising options for the treatment of photoaging.

Primary extramammary Paget disease (EMPD) is a rare skin disease of uncertain cellular derivation. Primary EMPD occurs in inguinal and genital skin in the absence of an associated internal malignancy. Several biomarkers have been implicated in EMPD, but the pathogenesis is debated. Serial skin biopsies were obtained from a 63-year-old male patient diagnosed with primary EMPD and treated with ingenol mebutate for 1 year. Although there was clinical improvement at 4 and 6 months, dermal invasion on pathology was noted at 1 year. Whole tissue transcriptome analysis during the treatment course revealed activation of mTOR and PI3K/AKT pathways, suggesting their involvement in the pathogenesis and invasion of EMPD. scRNA-seq was performed to define the EMPD cell cluster, compare the expression between normal and EMPD within epithelial clusters and evaluate the origin of the neoplastic cell using RNA velocity. With evidence of RNA velocity, we suggest Paget cells may arise from keratinocytes. Integration of scRNAseq and spatial transcriptomics via computational deconvolution methods revealed changes in the immune cell profile according to the time-interval along with treatment duration. Overall, these data inform on the pathomechanisms of EMPD formation and progression.

Speech recordings in call centers are narrowband and mixed with various noises. Developing a bandwidth expansion (BWE) model is important to mitigate the automated speech recognition (ASR) performance gap between the low and high sampling rate speech data. To further address the in-the-wild noise in call center settings, we propose an Embedding-Polished Wave-U-Net (EP-WUN) that includes an additional speech quality classifier to handle the noise and bandwidth expansion of 8k audio simultaneously. Our framework shows improved speech quality metrics on a well-known BWE dataset (Valentini-Botinhao corpus) when comparing to the current state-of-the-art noise-robust BWE model with 33% fewer parameters. It also achieves an 11.71% word error rate reduction when evaluating on a real-world interactive voice response system from the E.SUN bank.

This paper proposes to use both audio input and subject information to predict the personalized preference of two audio segments with the same content in different qualities. A siamese network is used to compare two inputs and predict the preference. Several different structures for each side of the siamese network are investigated. The baseline structure which uses only audio information involves using a pretrained audio encoder followed by fully connected layers. In several different proposed structures, the approach of concatenating subject information with audio embedding before feeding it into fully connected layers outperforms the baseline model the most, resulting in an increase in overall accuracy from 77.56% to 78.04%. Experimental results also demonstrate that utilizing the complete set of subject information, which includes age, gender, and headphone/earphone specifications such as impedance, frequency response range, and sensitivity, is more effective than using a subset of this information.