When the controlled-source electromagnetic (CSEM) data are contaminated by intense cultural noise and signal-to-noise ratio (S/N) is lower than 0 dB, existing denoising methods can hardly achieve good results. To overcome problem, a new strong-noise elimination method called inception-temporal convolutional network-shift-invariant sparse coding (IncepTCN-SISC) developed based on deep learning dictionary learning. First, novel neural network model IncepTCN created inception block temporal...

The magnetotelluric (MT) signals are susceptible to anthropogenic noise and the existing denoising methods have significant shortcomings in low-frequency situations. To address problem, we propose an innovative approach. It is different from that attempt achieve signal-noise separation through one step. process divided into two steps proposed effective dominant component high-frequency sequentially extracted deep learning dictionary learning. We a new network named DnCNN-GRU which combines...

We provide a speech coding scheme employing generative model based on SampleRNN that, while operating at significantly lower bitrates, matches or surpasses the perceptual quality of state-of-the-art classic wide-band codecs. Moreover, it is demonstrated that proposed can meaningful rate-distortion trade-off without retraining. evaluate in series listening tests and discuss limitations approach.

Abstract Magnetotelluric (MT) method is widely used for revealing deep electrical structure. However, natural MT signals are susceptible to cultural noises. In particular, the existing data-processing methods usually fail work when data contaminated by persistent or coherent To improve quality of collected with strong ambient noises, we propose a novel time-series editing based on improved shift-invariant sparse coding (ISISC), data-driven machine learning algorithm. First, redundant...

Magnetic object localization techniques have significant applications in automated surveillance and security systems, such as aviation aircrafts or underwater vehicles. In this letter, a practical algorithm was presented to determine the center coordinates magnetic moments of multiple objects using combination field vector its gradient tensor data. It formulates into nonlinear problem, which solved by Levenberg–Marquardt algorithm. The regularization parameters problem were adaptively varied...

The noise suppression method based on dictionary learning has shown great potential in magnetotelluric (MT) data processing. However, the constraints used existing algorithm’s need to set manually, which significantly limits its application. To solve this problem, we propose a deep optimized denoising method. We use convolutional network learn characteristic parameters of high-quality MT independently and then them as for so achieve fully adaptive sparse decomposition. uses unified all...

We propose a neural audio generative model, MDCTNet, operating in the perceptually weighted domain of an adaptive modified discrete cosine transform (MDCT). The architecture model captures correlations both time and frequency directions with recurrent layers (RNNs). An coding system is obtained by training MDCTNet on diverse set fullband monophonic signals at 48 kHz sampling, conditioned perceptual encoder. In subjective listening test ten excerpts chosen to be balanced across content types,...

Seismic data processing plays a pivotal role in extracting valuable subsurface information for various geophysical applications. However, seismic records often suffer from inherent random noise, which obscures meaningful geological features and reduces the reliability of interpretations. In recent years, deep learning methodologies have shown promising results performing noise attenuation tasks on data. this research, we propose modifications to standard U-Net structure by integrating dense...

Here we present a novel approach to conditioning the SampleRNN generative model for voice conversion (VC). Conventional methods VC modify perceived speaker identity by converting between source and target acoustic features. Our focuses on preserving content depends network learn style. We first train multi-speaker conditioned linguistic features, pitch contour, using speech corpus. Voice-converted is generated features contour extracted from speaker, identity. demonstrate that our system...

Music performance synthesis aims to synthesize a musical score into natural performance. In this paper, we borrow recent advances in text-to-speech and present the Deep Performer—a novel system for score-to-audio music synthesis. Unlike speech, often contains polyphony long notes. Hence, propose two new techniques handling polyphonic inputs providing fine-grained conditioning transformer encoder-decoder model. To train our proposed system, violin dataset consisting of paired recordings...

A closed-form formula is developed for the full magnetic gradient tensor of a polyhedral body with homogeneous magnetization vector. It based on direct derivative technique closed form field. These analytical expressions are implemented into an easy-to-use C++ package which simultaneously calculates potential, field, and targets. Modern unstructured tetrahedral grids adopted to represent so that our code can deal arbitrarily complicated prismatic tested verify accuracies formula. Excellent...

We consider source coding of audio signals with the help a generative model. use construction where waveform is first quantized, yielding finite bitrate representation. The then reconstructed by random sampling from model conditioned on quantized waveform. proposed scheme theoretically analyzed. Using SampleRNN as model, we demonstrate that structure provides performance competitive state-of-the-art tools for specific categories signals.

Object localization techniques have significant applications in civil fields and safety problems. A novel analytical formula is developed for accurate underwater aerial object real-time by combining gravitational field horizontal gradient anomalies. The proposed method enhances the accuracy of its excess mass estimation; it also effectively avoids possible numerical instability singularity previous works. Finally, a synthetic navigation model was adopted to verify performance. results show...

Zambia Copperbelt hosts numerous world‐class, high‐grade sediment‐hosted stratiform Cu deposits. The Lower Roan Group of the Katanga Supergroup is main ore host and mainly comprises clastic rocks. In this study, we conducted laser ablation inductively coupled plasma mass spectrometry detrital zircon dating on sedimentary rocks in Chambishi Basin, order to investigate Neoproterozoic rift evolution Congo Craton. Our results reveal that age population (97.1%) falls within 2.1–1.7 Ga (peaks at...

Abstract The existing sparse decomposition denoising methods for magnetotelluric (MT) data need to set the iterative stop condition manually, which not only has a large workload and high difficulty, but also easily causes subjective bias. To this end, we propose new adaptive representation method MT denoising. First, be processed is divided into high-quality segments noisy by machine learning algorithm. Then, characteristic parameters of are calculated, boundary value taken as threshold....