Search

This paper describes our NPU-ASLP system for the Audio-Visual Diarization and Recognition (AVDR) task in the Multi-modal Information based Speech Processing (MISP) 2022 Challenge. Specifically, the weighted prediction error (WPE) and guided source separation (GSS) techniques are used to reduce reverberation and generate clean signals for each single speaker first. Then, we explore the effectiveness of Branchformer and E-Branchformer based ASR systems. To better make use of the visual modality, a cross-attention based multi-modal fusion module is proposed, which explicitly learns the contextual relationship between different modalities. Experiments show that our system achieves a concatenated minimum-permutation character error rate (cpCER) of 28.13\% and 31.21\% on the Dev and Eval set, and obtains second place in the challenge., 2 pages, accepted by ICASSP 2023

The performance of the keyword spotting (KWS) system based on audio modality, commonly measured in false alarms and false rejects, degrades significantly under the far field and noisy conditions. Therefore, audio-visual keyword spotting, which leverages complementary relationships over multiple modalities, has recently gained much attention. However, current studies mainly focus on combining the exclusively learned representations of different modalities, instead of exploring the modal relationships during each respective modeling. In this paper, we propose a novel visual modality enhanced end-to-end KWS framework (VE-KWS), which fuses audio and visual modalities from two aspects. The first one is utilizing the speaker location information obtained from the lip region in videos to assist the training of multi-channel audio beamformer. By involving the beamformer as an audio enhancement module, the acoustic distortions, caused by the far field or noisy environments, could be significantly suppressed. The other one is conducting cross-attention between different modalities to capture the inter-modal relationships and help the representation learning of each modality. Experiments on the MSIP challenge corpus show that our proposed model achieves 2.79% false rejection rate and 2.95% false alarm rate on the Eval set, resulting in a new SOTA performance compared with the top-ranking systems in the ICASSP2022 MISP challenge., 5 pages. Accepted at ICASSP2023

Recently cross-channel attention, which better leverages multi-channel signals from microphone array, has shown promising results in the multi-party meeting scenario. Cross-channel attention focuses on either learning global correlations between sequences of different channels or exploiting fine-grained channel-wise information effectively at each time step. Considering the delay of microphone array receiving sound, we propose a multi-frame cross-channel attention, which models cross-channel information between adjacent frames to exploit the complementarity of both frame-wise and channel-wise knowledge. Besides, we also propose a multi-layer convolutional mechanism to fuse the multi-channel output and a channel masking strategy to combat the channel number mismatch problem between training and inference. Experiments on the AliMeeting, a real-world corpus, reveal that our proposed model outperforms single-channel model by 31.7\% and 37.0\% CER reduction on Eval and Test sets. Moreover, with comparable model parameters and training data, our proposed model achieves a new SOTA performance on the AliMeeting corpus, as compared with the top ranking systems in the ICASSP2022 M2MeT challenge, a recently held multi-channel multi-speaker ASR challenge., Accepted by SLT 2022

The recently proposed serialized output training (SOT) simplifies multi-talker automatic speech recognition (ASR) by generating speaker transcriptions separated by a special token. However, frequent speaker changes can make speaker change prediction difficult. To address this, we propose boundary-aware serialized output training (BA-SOT), which explicitly incorporates boundary knowledge into the decoder via a speaker change detection task and boundary constraint loss. We also introduce a two-stage connectionist temporal classification (CTC) strategy that incorporates token-level SOT CTC to restore temporal context information. Besides typical character error rate (CER), we introduce utterance-dependent character error rate (UD-CER) to further measure the precision of speaker change prediction. Compared to original SOT, BA-SOT reduces CER/UD-CER by 5.1%/14.0%, and leveraging a pre-trained ASR model for BA-SOT model initialization further reduces CER/UD-CER by 8.4%/19.9%., Comment: Accepted by INTERSPEECH 2023

In cooperative multi-agent reinforcement learning (MARL) settings, the centralized training with decentralized execution (CTDE) becomes customary recently due to the physical demand. However, the most dilemma is the inconsistency of jointly-trained policies and individually-optimized actions. In this work, we propose a novel value-based multi-objective learning approach, named Tchebycheff value decomposition optimization (TVDO), to overcome the above dilemma. In particular, a nonlinear Tchebycheff aggregation method is designed to transform the MARL task into multi-objective optimal counterpart by tightly constraining the upper bound of individual action-value bias. We theoretically prove that TVDO well satisfies the necessary and sufficient condition of individual global max (IGM) with no extra limitations, which exactly guarantees the consistency between the global and individual optimal action-value function. Empirically, in the climb and penalty game, we verify that TVDO represents precisely from global to individual value factorization with a guarantee of the policy consistency. Furthermore, we also evaluate TVDO in the challenging scenarios of StarCraft II micromanagement tasks, and extensive experiments demonstrate that TVDO achieves more competitive performances than several state-of-the-art MARL methods.

In this study, we propose a timbre-reserved adversarial attack approach for speaker identification (SID) to not only exploit the weakness of the SID model but also preserve the timbre of the target speaker in a black-box attack setting. Particularly, we generate timbre-reserved fake audio by adding an adversarial constraint during the training of the voice conversion model. Then, we leverage a pseudo-Siamese network architecture to learn from the black-box SID model constraining both intrinsic similarity and structural similarity simultaneously. The intrinsic similarity loss is to learn an intrinsic invariance, while the structural similarity loss is to ensure that the substitute SID model shares a similar decision boundary to the fixed black-box SID model. The substitute model can be used as a proxy to generate timbre-reserved fake audio for attacking. Experimental results on the Audio Deepfake Detection (ADD) challenge dataset indicate that the attack success rate of our proposed approach yields up to 60.58% and 55.38% in the white-box and black-box scenarios, respectively, and can deceive both human beings and machines., Comment: 5 pages

UniSpeech has achieved superior performance in cross-lingual automatic speech recognition (ASR) by explicitly aligning latent representations to phoneme units using multi-task self-supervised learning. While the learned representations transfer well from high-resource to low-resource languages, predicting words directly from these phonetic representations in downstream ASR is challenging. In this paper, we propose TranUSR, a two-stage model comprising a pre-trained UniData2vec and a phoneme-to-word Transcoder. Different from UniSpeech, UniData2vec replaces the quantized discrete representations with continuous and contextual representations from a teacher model for phonetically-aware pre-training. Then, Transcoder learns to translate phonemes to words with the aid of extra texts, enabling direct word generation. Experiments on Common Voice show that UniData2vec reduces PER by 5.3% compared to UniSpeech, while Transcoder yields a 14.4% WER reduction compared to grapheme fine-tuning., Comment: 5 pages, 3 figures. Accepted by INTERSPEECH 2023

General accent recognition (AR) models tend to directly extract low-level information from spectrums, which always significantly overfit on speakers or channels. Considering accent can be regarded as a series of shifts relative to native pronunciation, distinguishing accents will be an easier task with accent shift as input. But due to the lack of native utterance as an anchor, estimating the accent shift is difficult. In this paper, we propose linguistic-acoustic similarity based accent shift (LASAS) for AR tasks. For an accent speech utterance, after mapping the corresponding text vector to multiple accent-associated spaces as anchors, its accent shift could be estimated by the similarities between the acoustic embedding and those anchors. Then, we concatenate the accent shift with a dimension-reduced text vector to obtain a linguistic-acoustic bimodal representation. Compared with pure acoustic embedding, the bimodal representation is richer and more clear by taking full advantage of both linguistic and acoustic information, which can effectively improve AR performance. Experiments on Accented English Speech Recognition Challenge (AESRC) dataset show that our method achieves 77.42% accuracy on Test set, obtaining a 6.94% relative improvement over a competitive system in the challenge., Accepted by Interspeech 2022

The color parameters of untreated wood powder and wood powder after warm-press forming were measured according to standard colorimetric systems. The composition of the sample was tested by wet chemistry technique. The chemical compositions of the sample after warm-press forming were analyzed by using X-ray photoelectron spectroscopy (XPS) and UV–Vis. The results showed that as the forming temperature increased, the lignin, the extracts and the absorption in the visible range increased, while the cellulose and the O/C values decreased. Besides, the color of the sample became darker. The changes in the chemical composition on the surface of the wood powder compact induced the color change.

Introduction. Functional training of high-performance athletes is aimed at forming a holistic structure of functional support for special abilities of athletes, taking into account the sport, type of competition, specialization. Long-term functional training consists of stages, where the implementation of each stage is a condition for achieving the end result of a high level of functional support for special performance of athletes. Functional training of high-performance athletes is the final integrated component of long-term functional training. The aim of the study is to substantiate normative parameters of functional support of special working capacity of high-performance athletes in academic rowing. Material and methods. Oxygen consumption (VO2), CO2 emission level (VCO2), minute ventilation (VE) were determined for the cutaneous respiratory cycle using an Oxycon gas analyzer (Jaeger). The concentration of lactate in the blood was determined with an automatic biochemical analyzer Biosen. Results. As a result, the normative parameters of the functional support of rowers’ special working capacity are substantiated. Quantitative characteristics are designed for differentiated assessment of fast kinetics, steady state, fatigue compensation, as well as evaluation of these characteristics in the process of modeling the 2000 m distance. Indicators of differentiated assessment of men / women, respectively. Evaluation of rapid reaction kinetics: Lamax, mmol-l-1-9.6-11.8 / 8.5-10,0; EqPaCO2 u. o.- 3.5-4.0 / 2.4-3.1; EqCO2 y. at. - 30.0-34.2 / 27.9-29.9. Steady state of functions: VO2max, ml · min-1 · kg-1 - 68.5-72.9 / 62.1-66.5; EqCO2, y. at. - 32.2-36.6, EqO2, y. at. - 32.5-34.1 / 30.0-33.6; LaVO2max, mmol·l-1 - 13.9-15.9 / 11.8-13.7. Fatigue compensation: EqCO2 steady state / EqCO2 fatigue compensation period,% -7.8-9.9% / 5.5-6.7%; EqО2 steady state / EqО2 period of fatigue compensation,% - 3,6-5,7% / 3,1-4,9%; VO2max mean value for 30 from the period of fatigue compensation, ± 2.0 ml • min – 1 • kg – 1 68.0–74.1 / 59.5–65.7; La, mmol • l-1 - 18.0-21.2 / 15.6-17.9. The indicators recorded in the process of simulating of the 2000 m distance did not differ significantly

Speech data on the Internet are proliferating exponentially because of the emergence of social media, and the sharing of such personal data raises obvious security and privacy concerns. One solution to mitigate these concerns involves concealing speaker identities before sharing speech data, also referred to as speaker anonymization. In our previous work, we have developed an automatic speaker verification (ASV)-model-free anonymization framework to protect speaker privacy while preserving speech intelligibility. Although the framework ranked first place in VoicePrivacy 2022 challenge, the anonymization was imperfect, since the speaker distinguishability of the anonymized speech was deteriorated. To address this issue, in this paper, we directly model the formant distribution and fundamental frequency (F0) to represent speaker identity and anonymize the source speech by the uniformly scaling formant and F0. By directly scaling the formant and F0, the speaker distinguishability degradation of the anonymized speech caused by the introduction of other speakers is prevented. The experimental results demonstrate that our proposed framework can improve the speaker distinguishability and significantly outperforms our previous framework in voice distinctiveness. Furthermore, our proposed method also can trade off the privacy-utility by using different scaling factors., Comment: Submitted to ICASSP 2023

This paper presents the NWPU-ASLP speaker anonymization system for VoicePrivacy 2022 Challenge. Our submission does not involve additional Automatic Speaker Verification (ASV) model or x-vector pool. Our system consists of four modules, including feature extractor, acoustic model, anonymization module, and neural vocoder. First, the feature extractor extracts the Phonetic Posteriorgram (PPG) and pitch from the input speech signal. Then, we reserve a pseudo speaker ID from a speaker look-up table (LUT), which is subsequently fed into a speaker encoder to generate the pseudo speaker embedding that is not corresponding to any real speaker. To ensure the pseudo speaker is distinguishable, we further average the randomly selected speaker embedding and weighted concatenate it with the pseudo speaker embedding to generate the anonymized speaker embedding. Finally, the acoustic model outputs the anonymized mel-spectrogram from the anonymized speaker embedding and a modified version of HifiGAN transforms the mel-spectrogram into the anonymized speech waveform. Experimental results demonstrate the effectiveness of our proposed anonymization system., Comment: VoicePrivacy 2022 Challenge

Self-supervised speech pre-training empowers the model with the contextual structure inherent in the speech signal while self-supervised text pre-training empowers the model with linguistic information. Both of them are beneficial for downstream speech tasks such as ASR. However, the distinct pre-training objectives make it challenging to jointly optimize the speech and text representation in the same model. To solve this problem, we propose Text-Enhanced Self-Supervised Speech Pre-training (TESSP), aiming to incorporate the linguistic information into speech pre-training. Our model consists of three parts, i.e., a speech encoder, a text encoder and a shared encoder. The model takes unsupervised speech and text data as the input and leverages the common HuBERT and MLM losses respectively. We also propose phoneme up-sampling and representation swapping to enable joint modeling of the speech and text information. Specifically, to fix the length mismatching problem between speech and text data, we phonemize the text sequence and up-sample the phonemes with the alignment information extracted from a small set of supervised data. Moreover, to close the gap between the learned speech and text representations, we swap the text representation with the speech representation extracted by the respective private encoders according to the alignment information. Experiments on the Librispeech dataset shows the proposed TESSP model achieves more than 10% improvement compared with WavLM on the test-clean and test-other sets. We also evaluate our model on the SUPERB benchmark, showing our model has better performance on Phoneme Recognition, Acoustic Speech Recognition and Speech Translation compared with WavLM., Comment: 9 pages, 4 figures

The development of integrated energy systems helps to improve energy efficiency and meet actual energy demand of the users. For the energy consumption characteristics of the district combined heat and electricity (CHP) system, a method for modeling the CHP system with the participation of the whole heating process considering demand for heat load was proposed. First, the working principle of the unit equipment and network in the system is explained, and the corresponding mathematical model is given. Second, the model of each link is integrated under a unified framework, and the transmission characteristics of the multi-energy system are combined to obtain the overall control method of the system. Finally, a simple example is analyzed to verify the feasibility of the regulation model with Matlab/Simulink.

Continuous integrate-and-fire (CIF) based models, which use a soft and monotonic alignment mechanism, have been well applied in non-autoregressive (NAR) speech recognition with competitive performance compared with other NAR methods. However, such an alignment learning strategy may suffer from an erroneous acoustic boundary estimation, severely hindering the convergence speed as well as the system performance. In this paper, we propose a boundary and context aware training approach for CIF based NAR models. Firstly, the connectionist temporal classification (CTC) spike information is utilized to guide the learning of acoustic boundaries in the CIF. Besides, an additional contextual decoder is introduced behind the CIF decoder, aiming to capture the linguistic dependencies within a sentence. Finally, we adopt a recently proposed Conformer architecture to improve the capacity of acoustic modeling. Experiments on the open-source Mandarin AISHELL-1 corpus show that the proposed method achieves a comparable character error rates (CERs) of 4.9% with only 1/24 latency compared with a state-of-the-art autoregressive (AR) Conformer model. Futhermore, when evaluating on an internal 7500 hours Mandarin corpus, our model still outperforms other NAR methods and even reaches the AR Conformer model on a challenging real-world noisy test set., Comment: 5 pages,4 figures

Additional file 1. Supplementary material.

This paper proposes a new type of Spatial Modulation technology based on difference equations, which aims to explore a new anti-jamming transmission method in the airspace. The basic principle of the proposed system and the antenna mapping structure are introduced in this paper. The receiving algorithm is analyzed, and the performance difference with traditional Spatial Modulation is compared. Performance simulation shows that the system is slightly lower in anti-white noise performance than traditional Spatial Modulation technology, and the system performance can be improved by using a better antenna mapping structure. Also, the proposed system has a unique advantage in resisting intentional interference.

This paper creatively proposes a new Generalized Space Shift Keying (GSSK) mapping rule that can transmit different antenna numbers at the same time. The antenna combination and performance of GSSK are explored and compared. The transmitting end introduces the transmission principle and antenna combination rules of GSSK. The receiver uses the maximum likelihood algorithm to ensure that the antenna number can be correctly detected. The error problem of the antenna detection and demodulation link is clearly demonstrated. The performance of GSSK with different combinations is simulated and compared. Finally, the performance of GSSK combination is summarized.

The scroll compressors with asymmetrical suction chamber have large discharge loss, as the pressure in outside chamber is far larger than the discharge pressure when the discharge process begins. It is necessary to optimize the scroll profile around the discharge port to reduce the discharge loss. The theoretical method is to redesign the tip profile to obtain the same volume ratio of outside working chamber with the inner working chamber. However, as it is difficult to predict the discharge loss through the flank gap in discharge chamber, the optimum design cannot be obtained. In this paper, we establish an optimization platform based on the CFD method. The profile of orbiting scroll tip is parametrized with CAD software, then the CFD calculation will be done and the performance parameters will be extracted. The MOST optimization algorithm is used to search the best performance parameters and control the parameterized value of scroll profile. After iterations, the best profile with the modification angle of 66.3° was obtained. Compared to the original model, the volumetric efficiency of the compressor increased by 4.11%; and the isentropic efficiency increased by 4.91%. Afterwards, the flow field are analysed to study the characteristics of the optimized scroll compressor. The results show that the backflow, the secondary vortex in discharge chamber and the discharge mass flow pulsation were reduced, which led to the decrease of the discharge loss and the improvement of the isentropic efficiency.

Taking two (4 × 40) m continuous rail transit rigid bridges as practical examples, the train-SSFST (steel-spring floating slab track)-bridge system was established to acquire the vehicle induced vibration responses on SSFST. Based on the concept of combing cymbal PE-VEH and multilayer stacks PE-VEH together, a kind of PE-VEH featured small size, lightweight, high bearing capacity and excellent efficiency in energy harvesting is proposed. The prominent highlight of the designed PE-VEH is the force amplifying mechanism composed of cover plates and bending plates, which is of great help in improving the energy harvesting efficiency. Secondly, the 6 multilayer piezoelectric stacks are favorable to meet the requirements of small dimensions when bearing the vehicle induced vertical steel-spring fulcrum forces. Dynamic simulation results showed that the optimum positions for the designed PE-VEH to be connected in series with the steel-springs is at the end of the floating concrete slab (FCS). Comparing with existed low-frequency PE-VEHs through evaluating by indicators of power density and NPD, the excellent energy output efficiency of the designed PE-VEH is demonstrated, and influence on the practical train-SSFST-bridge system is negligible as well. For the example bridge, with 144 PE-VEHs arranged on 36 blocks of FCS, the total energy power obtained would reach up to 31.4 kJ, enough to satisfy the normal use of intermittent collection and transmission components for wireless BHM.

In view of the current status of the monitoring system for cable conduit protectors, the green environmental protection and the monitoring system for cable conduit protectors are closely combined to explore the construction of an omnidirectional, multi-level and wide-coverage intelligent cable conduit protector monitoring system to coordinate the advantages of advanced scientific research, technology, instruments and equipment. Based on the research background, the paper briefly introduces the high-temperature superconducting cable monitoring and protection system, and systematically explains the overall function of this research software, frame structure design, communication implementation method, various data record analysis and fixed value research.

To ensure the safety of power cables, temperature monitoring of power cables is critical. In view of the shortcomings of some optical fibre measurement products, a long-distance distributed optical fibre sensor is proposed, which has the advantages of high measurement sensitivity, fast response speed, low false alarm rate, large measurement length, and stable operation. The distributed optical fibre vibration sensing measurement equipment is used to monitor the vibration signals along the cable in real time, and the signal changes before and after the breakdown fault are visually displayed by the patio-temporal spectrum diagram, and the vibration signal of the breakdown point is analysed in time and frequency. The performance of the system was tested, and the experiment verified that the temperature measurement accuracy error range of the temperature measurement system is about ± 5 °C, and it has good stability. At the same time, the positioning accuracy is very high, and the fault point can be accurately located. This can avoid the occurrence of cable accidents.

To realize the self-learning operation modes recognition of air borne fire control radar, an optimized LSSVM (least squares support vector machine) algorithm based on grid search and K-fold cross validation is proposed. First, extract feature parameters of non-cooperative radar signal and establish library base of radar signal as training sample in LSSVM model. Next, apply grid search method in parameter optimization to realize model adjustment under the circumstance of uncertain sample range. In the sequel, use K-fold cross validation to realize performance evaluation and reduce model error caused by sample randomness, which improves generalization ability. Simulation results show that recognition accuracy of Velocity Search (VS) / Single Target Track (STT) / Range While Search (RWS) / Track While Scan (TWS) modes reaches 97%, proving good recognition performance and practical value of proposed method.

Regarding the deficiencies of clustering algorithm with fast method of searching and finding density peaks which is published in Science in 2014, an automatically and fast finding of clustering centers clustering algorithm is proposed, which applies linear regression and residual analysis and optimizes sample density value. The proposed algorithm improves location stability of clustering centers by measuring point density through sample's nearest neighbors information, and it determines clustering centers fast and automatically by applying linear regression and residual analysis, which reduces the subjectivity of artificial selection. Theoretical analysis and simulation results show that the proposed algorithm can overcome deficiencies of the original algorithm, what's more, the results of clustering effect and calculation time is superior to original algorithm, K-means and DBSCAN.

Path planning for jammers formation is vital to the success rate of penetration operations. Rasterization is applied to construct the discretized operation space. The natural terrains, including random surface model, mountain terrain model are built through the mathematical model. Then, the range threat, mountain threat, radar detection threat and aircraft flight constraint are established as the path threat factors. In view of the defects of the traditional PSO, GA is introduced to improve it through dividing the population into two parts to acquire the comprehensive advantages of the two algorithms. The simulation results indicate that the jamming path with low threat could be acquired through the planning with improved PSO-GA.

When input dimension of BP neural network increased, the structure of BP neural network will become especially complex and it is vulnerable to fall into local optimum in the training process. Therefore, a new algorithm of BP neural network optimized by rough set (RS) and mind evolutionary algorithm (MEA) is proposed. RS theory was used as the front-end date processing system to achieve attribute reduction, which played a major role in reducing the complexity of BP neural network and shortening the operation time. In order to optimize the performance and lower convergence speed of BP neural network, mind evolutionary algorithm (MEA) is applied to compute the global optimal initial weight and threshold of BP neural network. According to simulation experiment and contrasting with traditional BP neural network, the performance of modified BP neural network is improved, and the proposed algorithm is applicable and valid.

In order to improve the suction performance of the scroll compressors, five modification models are proposed by reducing or enlarging the area of the suction flow passage based on a scroll refrigeration compressor. The three-dimensional numerical simulation model is established using the commercial software PumpLinx to calculate the flow field and the performance of every modified model. The comparison and analyses has been done based on the simulation results. The results indicated that the enlargement of the suction passage only improve the volumetric efficiency a little while the shrink of the suction passage can improve it effectively. The mass flow rate pulsation when the modification angle is 270° is much higher than that of the original model, but it decreases with the modification angle. As the modification angle increases, the volume of inner suction passage and the back flow rate from the inner suction passage decrease and less fluid from suction chamber 2 flows back to the suction port, so the average mass flow rate increases and the mass flow rate pulsation decreases. The Model 5 has the maximum volumetric efficiency and lower mass flow pulsation, but the tangential and radial forces on the orbiting scroll are the highest.

Radial resistant force against the moving tool was measured dynamically and influences of some factors on the force were analyzed.It shows that the force rises up sharply at the beginning of friction stir welding As soon as reaching the top it decreases down rapidly,and then goes to a value at which the welding procedure is steady.If increasing the rotation speed of tool,but keeping all other parameters and factors fLxed,the top force and the steady force are both decreased,and the times taken for reaching the top force and steady force are shortened. If increasing welding speed only,the top force goes up greatly, and the steady force increases or decreases in a rather small range,and the time taken for the top force changes a little,but the time for the steady force is prolonged;The influences of the diameter of pin on the force are complex,because the change of the diameter of pin leads to the variation of frictional heating power and the cooling power simultaneously.