The growing demand for psychological support underscores the lack of high-quality counseling dialogue datasets, particularly in non-English contexts. We propose PGSim, a Path-Guided Simulation framework that mirrors real counseling processes—symptom description, problem identification, cause analysis, strategy planning, and iterative adjustment. PGSim models each user scenario as a fine-grained quadruple {Group, Psychological Problem, Problem Cause, Support Focus} and guides dialogue generation through expert-annotated strategy paths. Real counseling dialogues and expert-edited samples are used to fine-tune two language models: a Dialog Generator for strategy-aligned dialogue creation and a Dialog Modifier for expert-level refinement. After automated and human verification, we construct the Chinese Psychological support Dialogue Dataset (CPsDD), containing 68K dialogues across 13 groups, 16 problems, 13 causes, and 12 support focuses. We further present the Comprehensive Agent Dialogue Support System (CADSS), which integrates profiling, summarization, strategy planning, and empathetic response. Experiments on CPsDD and ESConv demonstrate that CADSS achieves state-of-the-art results on Strategy Prediction and Emotional Support Conversation tasks.

This brief presents an amplifierless current-to-digital converter (CDC) that uniquely integrates an open-loop pseudo-differential current mirror with a current-integration successive-approximation-register analog-to-digital converter (ADC). The proposed architecture enables the CDC to achieve high-speed operation at low power consumption, which is critical for the intended applications in dynamic optical coherence tomography (OCT) systems. Fabricated in 65-nm CMOS, the prototype occupies 0.019 mm2, consumes$380~\mu $W from a 1-V supply, and achieves a 47-dB dynamic range (DR) with a 50-MS/s sample rate. It achieves Walden’s and Schreier’s figures of merit of 92 fJ/step and 148 dB, respectively, both being the best among reported CDCs.

Table -to-text generation aims to automatically produce natural language descriptions from structured or semi-structured tabular data.Unlike traditional text generation tasks, it requires models to accurately understand and represent table structures.Existing approaches typically process tables by linearizing them or converting them into graph structures.However, these methods either fail to adequately capture the table structure or rely on complex attention mechanisms, limiting their applicability.To tackle these challenges, we propose QuASAR, a question-driven self-supervised approach designed to enhance the model's structural perception and representation capabilities.Specifically, QuASAR formulates a set of structure-related queries for self-supervised training, explicitly guiding the model to capture both local and global table structures.Additionally, we introduce two auxiliary pre-training tasks: a word-to-sentence reconstruction task and a numerical summarization task, which further enhance the fluency and factuality of the generated text.Experimental results on the ToTTo and HiTab datasets demonstrate that our approach produces higher-quality text compared to existing methods.

This paper aims to introduce a robust singing voice synthesis (SVS) system to produce very natural and realistic singing voices efficiently by leveraging the adversarial training strategy. On one hand, we designed simple but generic random area conditional discriminators to help supervise the acoustic model, which can effectively avoid the over-smoothed spectrogram prediction and improve the expressiveness of SVS. On the other hand, we subtly combined the spectrogram with the frame-level linearly-interpolated F0 sequence as the input for the neural vocoder, which is then optimized with the help of multiple adversarial conditional discriminators in the waveform domain and multi-scale distance functions in the frequency do-main. The experimental results and ablation studies concluded that, compared with our previous auto-regressive work, our new system can produce high-quality singing voices efficiently by fine-tuning different singing datasets covering from several minutes to a few hours. A large number of synthesized songs with different timbres are available online1and we highly recommend readers to listen to them.

We propose a novel method for zero-shot cross-lingual TTS task by using multi-stream text encoder and efficient speaker representation. Specifically, a unified multi-stream text encoder that takes both advantages of Transformer and CBHG is proposed to retain multiple hypotheses about input representations. For Transformer based stream, a multi-stream Transformer is further proposed to strengthen these hypotheses. Then the speaker representations are extracted from audio signals by a speaker encoder with a random sampling mechanism and a language adversarial loss, aiming to extract speaker embedding features that are independent of both content information and language identity. Meanwhile, we propose an efficient zero-shot cross-lingual transfer strategy with the help of other target lingual speakers’ data and a language-balanced sampling strategy. The Experimental results show the proposed method not only could achieve higher speech quality and speaker similarity (with an average absolute improvement of 0.38 and 0.27 in MOS respectively) for zero-shot cross-lingual transfer, but also helpful for few-shot cross-lingual transfer in which has multi-lingual data.1

In this paper, we develop a new multi-singer Chinese neural singing voice synthesis (SVS) system named WeSinger. To improve the accuracy and naturalness of synthesized singing voice, we design several specifical modules and techniques: 1) A deep bi-directional LSTM-based duration model with multi-scale rhythm loss and post-processing step; 2) A Transformer-alike acoustic model with progressive pitch-weighted decoder loss; 3) a 24 kHz pitch-aware LPCNet neural vocoder to produce high-quality singing waveforms; 4) A novel data augmentation method with multi-singer pre-training for stronger robustness and naturalness. To our knowledge, WeSinger is the first SVS system to adopt 24 kHz LPCNet and multi-singer pre-training simultaneously. Both quantitative and qualitative evaluation results demonstrate the effectiveness of WeSinger in terms of accuracy and naturalness, and WeSinger achieves state-of-the-art performance on the recent public Chinese singing corpus Opencpop\footnote{https://wenet.org.cn/opencpop/}. Some synthesized singing samples are available online\footnote{https://zzw922cn.github.io/wesinger/}.

In this paper, we propose a novel method for fast and efficient few-shot TTS task, which is able to disentangle linguistic and speaker representations. Specifically, an adversarial training strategy is firstly employed to wipe out speaker information from the linguistic representations. Then the speaker representations are extracted from audio signals by a speaker encoder with a random sampling mechanism and a speaker classifier, aiming to extract speaker embedding features that are independent of content information (such as prosody and style etc). Meanwhile, for faster and efficient adaptation, we further introduce the prior alignment knowledge between the text and audio pairs and propose a multi-alignment guided attention to help the attention learning. The Experimental results show the proposed method not only could generate higher speech quality and speaker similarity with an average absolute improvement of 0.26 and 0.30 in MOS respectively, when adapting to new speakers with 20 utterances, but also converge much faster and efficient. More-over, we can achieve a MOS of 4.45 for a premium voice, which outperforms a single speaker model of 4.23.1

A frame-unit-selection based voice conversion system proposed earlier by us is revisited here to enhance its performance in both speech naturalness and speaker similarity. Speaker independent, bilingual (Mandarin Chinese and American English) deep neural net (DNN) acoustic model’s output, frame-level phone posterior probability (PPP), is used to represent the phonetic information. The corresponding frame-level F0 is used as the prosodic information. Kullback-Leibler divergence (KLD) between source and target PPPs (phonetic distortion) and the absolute difference between normalized source and target F0 (prosodic distortion) are used for selecting target frame candidates to construct a search lattice. The optimal target unit trajectory is obtained by Viterbi algorithm which tries to minimize the dynamic acoustic difference between the acoustic trajectory of the source speech and target candidates. The obtained spectral trajectory together with the enhanced pitch period and pitch correlation trajectory are sent to LPCNet vocoder to synthesize the converted waveforms. Compared with the top-rank system in Voice Conversion Challenge 2018, our new system can achieve on-par performance on studio to studio American English VC test, and better performance on non-studio to studio Mandarin Chinese VC test, in both speech naturalness MOS and speaker similarity DMOS.

Although Transformer based neural end-to-end TTS model has demonstrated extreme effectiveness in capturing long-term dependencies and achieved state-of-the-art performance, it still suffers from two problems. 1) limited ability to model sequential and local structures in sequences; 2) heavily rely on position embeddings that have limited effect but require an amount of design efforts. In this paper, we introduce local recurrent neural network (Local-RNN) into Transformer to make full use of the advantages of both RNN and Transformer while mitigating their drawbacks. The sequential and local structures could be effectively modeled by Local-RNN, while the long-term dependencies could be captured by Transformer without any use of position embeddings. Subjective evaluation results show our proposed model outperforms baseline (Transformer) with a gap of 0.12 in MOS and achieves close to human quality (4.34 vs. 4.45 in MOS) on general test. Case level intelligibility test also show an absolute improvement of 6.5% in case level intelligibility rate over the baseline on a challenging test.

This paper presents an architecture to perform speaker adaption in long short-term memory (LSTM) based Mandarin statistical parametric speech synthesis system. Compared with the conventional methods that focused on using fixed global speaker representations in utterance level for speaker recognition task, the proposed method extracts speaker representations in utterance and phoneme level, which can describe more pronunciation characteristics in phoneme level. And an attention mechanism is deployed to combine each level representations dynamically to train a task-specific phoneme dependent speaker embedding. To handle the unbalanced database and avoid over-fitting, the model is factored into an average model and an adaptation model and combined by an attention mechanism. We investigate the performance of speaker representations extracted by different methods. Experimental results confirm the adaptability of our proposed speaker embedding and model factorization structure. And listening tests demonstrate that our proposed method can achieve better adaptation performance than baselines in terms of naturalness and speaker similarity.

Neural end-to-end TTS can generate very high-quality synthesized speech, and even close to human recording within similar domain text. However, it performs unsatisfactory when scaling it to challenging test sets. One concern is that the encoder-decoder with attention-based network adopts autoregressive generative sequence model with the limitation of exposure bias To address this issue, we propose two novel methods, which learn to predict future by improving agreement between forward and backward decoding sequence. The first one is achieved by introducing divergence regularization terms into model training objective to reduce the mismatch between two directional models, namely L2R and R2L (which generates targets from left-to-right and right-to-left, respectively). While the second one operates on decoder-level and exploits the future information during decoding. In addition, we employ a joint training strategy to allow forward and backward decoding to improve each other in an interactive process. Experimental results show our proposed methods especially the second one (bidirectional decoder regularization), leads a significantly improvement on both robustness and overall naturalness, as outperforming baseline (the revised version of Tacotron2) with a MOS gap of 0.14 in a challenging test, and achieving close to human quality (4.42 vs. 4.49 in MOS) on general test.

Neural end-to-end TTS such as Tacotron like network can generate very high-quality synthesized speech, and even close to human recording for similar domain text. However, it performs unsatisfactory when scaling it to some challenging test sets. One concern is that the encoder-decoder with attention-based network adopts autoregressive generative sequence model with the limitation of “exposure bias”: errors made early could be quickly amplified, harming subsequent sequence generation. To address this issue, we propose two novel methods, which aim at predicting future by improving the agreement between forward and backward decoding sequence. The first one (denoted as MRBA) is achieved by adding divergence regularization terms to model training objective to maximize the agreement between two directional models, namely L2R (which generates targets from left-to-right) and R2L (which generates targets from right-to-left). While the second one (denoted as BDR) operates on decoder-level and exploits the future information during decoding. By introducing regularization term into the training objective of forward-backward decoders, the forward-decoder's hidden states are forced to be close to the backward-decoder's. Thus, the hidden representations of a unidirectional decoder are encouraged to embed some useful information about the future. Moreover, in order to make forward and backward decoding to improve each other in an interactive process, a joint training method is designed. Experimental results on both English and Mandarin dataset show that our proposed methods especially the second one (BDR), lead to a significantly improvement on both robustness and overall naturalness, as achieving obvious preference advantages in a challenging test, and achieving state-of-the-art performance (outperforming baseline “the revised version of Tacotron2” with a gap of 0.13 and 0.12 for English and Mandarin in MOS, respectively) on a general test.

Pitch is an important characteristic of speech and is useful for many applications. However, it is still challenging to estimate pitch in strong noise. In this paper, we propose a joint training approach to determinate pitch. First, a Bidirectional Long Short-Term Memory Recurrent Neural Networks (BLSTMRNN) is trained to map the noisy to clean speech features. Second, the pitch estimation is also a BLSTM-RNN model. The feature mapping neural network serves as a noise normalization module aiming at explicitly generating the clean features which are easier to estimate pitch by the following neural network. BLSTM-RNN is trained on sequential frame-level features and capable of learning temporal dynamics. We also propose to take into account bottleneck features for pitch estimation. The experimental results show that the proposed method can obtain accurate pitch estimation and they show good generalization ability to new speakers and noisy conditions. The proposed approach also significantly outperforms other state-of-the-art pitch estimation algorithms.

In this paper, we generalize the widely used simultaneous block iterative reconstruction algorithm and show that it converges, at a linear rate, to a weighted least-squares and weighted minimum-norm reconstruction. Our theoretical result provides a much simpler proof of the convergence properties obtained by Jiang and Wang and covers a much more general class of algorithms. The frequency domain iterative reconstruction algorithm is then introduced as a special application of our theory.

The design of beamformer is a critical component in the development of ultrasound imaging. Conventional delay-and-sum beamformer is intuitive and easy to implement, however the range resolution is limited by the pulse bandwidth. In this paper, we propose a technique that optimizes the reconstruction kernel to approximate an ideal impulse, while also controlling the mainlobe/sidelobe tradeoff. Preliminary simulation has demonstrated that the new method achieves significantly better range resolution than simple delay-and-sum beamformers for medium bandwidth near-field imaging.

To resolve fine details, or to accurately segment tumors from background activities, it is desired that the reconstruction image of the SPECT may preserve the edges and achieve high resolution. In this paper, we develop a blind deblurring reconstruction technique estimate of both the actual image and the PSF of the system, and enhance the performance of iterative reconstruction using this technique. A blurred SPECT reconstruction can be viewed as the convolution of a low-pass PSF with the actual image, where both the PSF and the actual image are unknown in practice. The PSF of a SPECT system is determined by the combined effect of several factors, which include the gamma camera PSF, the scattering, and pinhole PSF for pinhole SPECT systems, and therefore is hard to be determined analytically. Inspired by the blind deconvolution algorithm, we formulate a blind deblurring reconstruction algorithm, which also consists of two iterative update sequences, which are corresponded for the PSF and the SPECT reconstruction, respectively. In the phantom study, the algorithm reduces image blurring and preserves the edges without introducing extra artifacts. The localized measurement shows that the performance of reconstruction image improved by up to 50%. In experimental studies, the contrast and quality of reconstruction is substantially improved. Therefore, algorithm shows promising in tumor localization and quantification.

We introduce general linear pre- and post- backprojection filters into the widely used simultaneous block iterative reconstruction algorithm, and show that the generalized algorithm converges, at linear speed, to a weighted least squares weighted minimum norm reconstruction, where the weighting matrices are determined by the filters. Our theoretical results expand several existing algorithms and open the possibility of selecting new filters. Examples of the application of our new theory are given.

The reconstruction of 3D ultrasound (US) images from mechanically registered but otherwise irregularly positioned B-scan slices is of great interest in image guided therapy procedures, such as that used for the treatment of prostate cancer. The conventional reconstruction method simply interpolates slices from the same angle and then compounds the interpolated data from different angles. This method results in spatial resolution far inferior to the in-plane resolution of each B-scan slice. We propose a novel reconstruction method which properly weights the frequency components of data sets from different angles in a Wiener filter/MMSE fashion. Simulation results for synthetic US images are presented to demonstrate the excellent reconstruction.

Magnetic resonance imaging (MRI) requires reconstruction of an image from non-uniformly sampled Fourier domain data. The point spread function (PSF) is strongly affected by the Fourier weights, also called density compensation function (DCF). We formulate the DCF selection as an optimization problem with linear matrix inequality (LMI) constraints. The maximum sidelobe level of the PSF is minimized while the mainlobe width is held constant. Our approach is in contrast to existing suboptimal approaches where a DCF is first found based on local sampling density, and then a secondary windowing function is applied to reduce sidelobes. Reconstructions of simulated data demonstrate that our approach produces more accurate and visually better looking images.

The problem of extracting time resolved beat spacing intervals from atrial fibrillation (AF) electrograms using wavelet analysis techniques is considered. The problem has the important application of localizing organized but intermittent sources of AF from complex time-dependent AF electrograms measured with a basket catheter. Analysis of synthesized electrograms demonstrates that beat spacing can be extracted accurately. The technique is then applied to acute and chronic AF electrograms measured from canine models of AF and comparisons of the results are examined.

The problem of retrieving spiral wave parameters (frequency, radial velocity, and center location) using a minimal number (4) of spatial sensors is considered. The problem has the important application of localization of spiral wave sources of atrial fibrillation from basket catheter electrograms. Numerical simulations demonstrate that our algorithm works effectively for a wide range of parameters, and for spiral waves generated by a cellular automaton model of cardiac wave propagation.

A novel parametric algorithm that can retrieve roughly 0.25MN 2D exponentials or 0.343MN 2D harmonics from M/spl times/N array data is presented. This compares favorably with most existing algorithms which can retrieve only order max(M, N) exponentials or harmonics. The algorithm is not Fourier resolution limited, and requires neither searching in 2D space nor 2D polynomial rooting. A specific example of retrieving 4 harmonics from a 3/spl times/3 array is developed in detail and numerical performance is demonstrated.

This work presents a rigorous mathematical derivation of an effective approximate solution to the near-field 3-D inverse scattering/imaging problem for a system where the coherent signals are transmitted, and the scattered signals are subsequently received at individual transmitters and receivers. Potential applications of this technology include a variety of sub-surface imaging modalities such as: UHF foliage penetrating SAR, ground penetrating radar for land mine detection, and electromagnetic millimeter-wave scanning for concealed weapon detection.

We propose an optically (though not all-optically) decodable error-correction code. The structure of the proposed code is defined in the Fourier spectrum domain. The Fourier spectrum of the codewords has zeros. Thus, we can view any signal appearing in these spectrum zeros as the "syndrome", since the codewords have no energy in these spectrum zeros. The optical technique of the Fourier transform is used for fast decoding speed. The proposed code is nonlinear and of length eight. This code has a total of eight codewords (so the code rate is (log/sub 2/8)/8=3/8) and can correct 1.75 consecutive bit errors on average.

We propose a fuzzy PIC multiuser detection/vector channel prediction scheme in Rayleigh fading channels. The vector channel prediction is based on the first-order auto-regressive model and the expectation-maximization algorithm. The signal-to-interference ratio and signal-to-noise ratio are estimated from the vector channel model's parameters, and we adapt the weight of each interference cancellation path via fuzzy inference mechanism. The proposed fuzzy PIC and vector channel prediction cooperate in a way that fuzzy PIC' some input parameters come from the channel predictor and fuzzy PIC makes the channel predictor more accurate at the next stage. Computing weights via the fuzzy adaptive method adds insignificant complexity because it involves only table lookup. The overall complexity is still linear in the number of users. The simulation results show that the proposed fuzzy PIC/vector channel prediction scheme performs better than previous improved PIC schemes in Rayleigh fading channels.

An interpolation problem that is important in viral X-ray crystallography is considered. The problem requires new methods because (1) the function is known to have icosahedral symmetry, (2) the data is corrupted by experimental errors and therefore lacks the symmetry, (3) the problem is 3D, (4) the measurements are irregularly spaced, and (5) the number of measurements is large (10**4). A least-squares approach is taken using two sets of basis functions: the functions implied by a minimum-energy band-limited exact interpolation problem and a complete orthonormal set of band-limited functions. A numerical example on Cowpea Mosaic virus is described.

An object model and estimation procedure for three-dimensional (3-D) reconstruction of objects from measurements of the spherically averaged Fourier transform magnitudes is described. The motivating application is the 3-D reconstruction of viruses based on solution X-ray scattering data. The object model includes symmetry, positivity and support constraints and has the form of a truncated orthonormal expansion and the parameters are estimated by maximum likelihood methods. Successful 3-D reconstructions based on synthetic and experimental measurements from Cowpea mosaic virus are described.

A model-based method for reconstructing the 3-dimensional structure of icosahedrally-symmetric viruses from solution X-ray scattering is presented. The algorithm is an iterative algorithm that is a generalization of the Gerchberg-Saxton algorithm from 3 dimensions (measure the magnitude squared of an unknown complex number) to n dimensions (measure the norm squared of an unknown n-vector). An example of the reconstruction, for data from cowpea mosaic virus, is described. The major opportunity provided by solution X-ray scattering is the ability to study the dynamics of virus particles in solution, information that is not accessible to crystal X-ray diffraction experiments.