Search

We introduce LyricWhiz, a robust, multilingual, and zero-shot automatic lyrics transcription method achieving state-of-the-art performance on various lyrics transcription datasets, even in challenging genres such as rock and metal. Our novel, training-free approach utilizes Whisper, a weakly supervised robust speech recognition model, and GPT-4, today's most performant chat-based large language model. In the proposed method, Whisper functions as the "ear" by transcribing the audio, while GPT-4 serves as the "brain," acting as an annotator with a strong performance for contextualized output selection and correction. Our experiments show that LyricWhiz significantly reduces Word Error Rate compared to existing methods in English and can effectively transcribe lyrics across multiple languages. Furthermore, we use LyricWhiz to create the first publicly available, large-scale, multilingual lyrics transcription dataset with a CC-BY-NC-SA copyright license, based on MTG-Jamendo, and offer a human-annotated subset for noise level estimation and evaluation. We anticipate that our proposed method and dataset will advance the development of multilingual lyrics transcription, a challenging and emerging task., 9 pages, 2 figures, 5 tables, accepted by ISMIR 2023

In the era of extensive intersection between art and Artificial Intelligence (AI), such as image generation and fiction co-creation, AI for music remains relatively nascent, particularly in music understanding. This is evident in the limited work on deep music representations, the scarcity of large-scale datasets, and the absence of a universal and community-driven benchmark. To address this issue, we introduce the Music Audio Representation Benchmark for universaL Evaluation, termed MARBLE. It aims to provide a benchmark for various Music Information Retrieval (MIR) tasks by defining a comprehensive taxonomy with four hierarchy levels, including acoustic, performance, score, and high-level description. We then establish a unified protocol based on 14 tasks on 8 public-available datasets, providing a fair and standard assessment of representations of all open-sourced pre-trained models developed on music recordings as baselines. Besides, MARBLE offers an easy-to-use, extendable, and reproducible suite for the community, with a clear statement on copyright issues on datasets. Results suggest recently proposed large-scale pre-trained musical language models perform the best in most tasks, with room for further improvement. The leaderboard and toolkit repository are published at https://marble-bm.shef.ac.uk to promote future music AI research.

Self-supervised learning (SSL) has shown promising results in various speech and natural language processing applications. However, its efficacy in music information retrieval (MIR) still remains largely unexplored. While previous SSL models pre-trained on music recordings may have been mostly closed-sourced, recent speech models such as wav2vec2.0 have shown promise in music modelling. Nevertheless, research exploring the effectiveness of applying speech SSL models to music recordings has been limited. We explore the music adaption of SSL with two distinctive speech-related models, data2vec1.0 and Hubert, and refer to them as music2vec and musicHuBERT, respectively. We train $12$ SSL models with 95M parameters under various pre-training configurations and systematically evaluate the MIR task performances with 13 different MIR tasks. Our findings suggest that training with music data can generally improve performance on MIR tasks, even when models are trained using paradigms designed for speech. However, we identify the limitations of such existing speech-oriented designs, especially in modelling polyphonic information. Based on the experimental results, empirical suggestions are also given for designing future musical SSL strategies and paradigms.

While end-to-end systems are becoming popular in auditory signal processing including automatic music tagging, models using raw audio as input needs a large amount of data and computational resources without domain knowledge. Inspired by the fact that temporal modulation is regarded as an essential component in auditory perception, we introduce the Temporal Modulation Neural Network (TMNN) that combines Mel-like data-driven front ends and temporal modulation filters with a simple ResNet back end. The structure includes a set of temporal modulation filters to capture long-term patterns in all frequency channels. Experimental results show that the proposed front ends surpass state-of-the-art (SOTA) methods on the MagnaTagATune dataset in automatic music tagging, and they are also helpful for keyword spotting on speech commands. Moreover, the model performance for each tag suggests that genre or instrument tags with complex rhythm and mood tags can especially be improved with temporal modulation., Comment: Submitted to ICASSP 2023

The traditional method of searching transmission section generally has the problem of omission and time-consuming. A topology converging algorithm of power grid adjacency matrix is proposed in this paper. It aggregates neighboring buses based on a simple matrix transformation rule, and identify key transmission sections based on matrix operations and power flow distribution factors. This method has the advantage of no presetting the sub-zone, which can avoid the simplification of the grid topology in the sub-zone and improve the calculation accuracy. The complex logical operations of the traversal algorithm adopted by most of traditional methods can be avoided, and the calculation speed can be effectively improved. Besides, the algorithm also has a memory characteristic which can be applied to the online search of key transmission sections. The results of several IEEE test systems show the effectiveness of the proposed algorithm.

Coral mud is a form of calcareous ooze composed of fine-grained coral debris distributed as interlayers due to particle sorting during the formation of reclamation reefs in the South China Sea. To analyze the effect of coral mud on the stability of reef foundations formed by dredger fill, in this paper, the effects of initial concentrations, initial settlement heights, and salinity levels on the sedimentary characteristics of coral mud are discussed based on laboratory simulation tests. Comparative tests of marine sedimentary soil with similar particle compositions and of terrestrial silty clay with similar plastic indexes were carried out, and the sedimentation velocity characteristics of coral mud were identified at the microscopic level. The experimental results show that reducing initial concentrations and initial settlement heights are conducive to the coral mud filling process and that a medium salt content has no significant effect. The sedimentation rate of coral mud is roughly 3–7 times that of common soil, and the time spent in the isokinetic sedimentation stage is only 1/3 that of common soil. Microscopic studies show that the settling velocity of clay in water is proportional to the sphericity of flocs. According to our experimental results, the relationship between the sedimentation velocity and initial concentrations of coral mud is obtained and can be used to guide the design of similar calcareous soil projects across the globe and of dredger filling projects in particular.

Despite recent achievements of deep learning automatic music generation algorithms, few approaches have been proposed to evaluate whether a single-track music excerpt is composed by automatons or Homo sapiens. To tackle this problem, we apply a masked language model based on ALBERT for composers classification. The aim is to obtain a model that can suggest the probability a MIDI clip might be composed condition on the auto-generation hypothesis, and which is trained with only AI-composed single-track MIDI. In this paper, the amount of parameters is reduced, two methods on data augmentation are proposed as well as a refined loss function to prevent overfitting. The experiment results show our model ranks $3^{rd}$ in all the $7$ teams in the data challenge in CSMT(2020). Furthermore, this inspiring method could be spread to other music information retrieval tasks that are based on a small dataset., Comment: submitted to CSMT 2020 challenge track

Soil voltage is generally assumed to show a linear relationship with distance from the cathode according to the established electroosmotic consolidation equation. However, this assumption is inconsistent with experimental results. To more reasonably reflect the soil consolidation process during electroosmosis treatment, it is necessary to consider the influence of the actual soil voltage distribution trend when establishing the electroosmotic consolidation equation. Electroosmosis results show that soil voltage exhibits nonlinear distribution characteristics against distance from the cathode. The change trend of soil voltage can be well reflected by cubic polynomial fitting. Then, the anodic electrode was taken as the research object, and a two-dimensional horizontal plane model of electroosmosis was established because it represents practical electroosmosis applications more closely than some other models. Based on this established model, the dissipation equation describing the excess pore water pressure and the soil consolidation equation were derived for the electroosmosis treatment process. The derivation process considered both linear and nonlinear soil voltage distributions, wherein the anode was closed and the cathode was open. Finally, the analytical solution was analyzed and validated with model test cases in terms of the excess pore water pressure and average moisture content of the soil. The trend observed in the measured excess pore water pressure was more consistent with that of the theoretical results calculated assuming a nonlinear soil voltage distribution than that obtained using a linear distribution. In addition, the measured values of the average moisture content in the soil were closer to the values calculated under a nonlinear distribution of soil voltage than to those calculated under a linear distribution. These results further show that the established consolidation equation is reasonable when a nonlinear distribution of soil voltage is considered. The proposed consolidation equation can thus improve the application of electroosmotic methods in the future.

The growing penetration of wind power in a power system brings great challenges to system operation flexibility. For generation planning in presence of high wind power penetration, it is essential to take the operation flexibility of the system into account. Firstly, this paper developed the system operation flexibility metrics through considering the flexibility contribution of thermal generating units (TGUs) by operational state transition. Secondly, a planning model for the bundled wind-thermal-storage generation system (BWTSGS) that considers the operation flexibility constraints is proposed. The planning model is used to determine the power and energy rating of an energy storage system (ESS) as well as the type and number of TGUs. A daily scheduling simulation model of a BWTSGS is proposed to calculate the operation cost for the planning model and consider the sequential operation characteristics of the BWTSGS. Further, in order to accelerate the computation, a wind power sequential clustering technique based on the discrete Fourier transform (DFT) method is developed for improving the computational efficiency. Case studies have been conducted on a 1000-MW wind farm to demonstrate the validity and effectiveness of the proposed model.

After reform of non-tradable shares to complete, along with limits sells the stock in batches going on the market, the increase of tradable shares proportion has the remarkable influence on liquidity level in China A shares market, thus also changed the liquidity premium. This paper uses panel data model, selects two liquidity variable from different angle, tests and analyzes the liquidity premium before and after the reform. The result shows that: the liquidity premium exist in market before and after the reform; because of the market liquidity condition improving, negative effect between illiquidity and expected return strengthen, namely stocks with low liquidity have higher expected return.

The authors took the stomatology as an application area for using VR techniques in training manual skills. In order to get real data for the shape of the oral cavity, we took complete dentitions from an young male corpse. While keeping their original geometrical shapes and relations to each other intact, we implemented a virtual oral environment with a set of virtual dental operations on this model. The environment was implemented in a style of agent oriented programming. This environment was used in an ICAI system for training stomatology skills.