Relative Error Rate Research Articles

A Phonetic-Semantic Pre-Training Model for Robust Speech Recognition

Robustness is a long-standing challenge for automatic speech recognition (ASR) as the applied environment of any ASR system faces much noisier speech samples than clean training corpora. However, it is impractical to annotate every types of noisy environments. In this work, we propose a novel phonetic-semantic pre-training (PSP) framework that allows a model to effectively improve the performance of ASR against practical noisy environments via seamlessly integrating pre-training, self-supervised learning, and fine-tuning. In particular, there are three fundamental stages in PSP. First, pre-train the phone-to-word transducer (PWT) to map the generated phone sequence to the target text using only unpaired text data; second, continue training the PWT on more complex data generated from an empirical phone-perturbation heuristic, in additional to self-supervised signals by recovering the tainted phones; and third, fine-tune the resultant PWT with real world speech data. We perform experiments on two real-life datasets collected from industrial scenarios and synthetic noisy datasets, which show that the PSP effectively improves the traditional ASR pipeline with relative character error rate (CER) reductions of 28.63% and 26.38%, respectively, in two real-life datasets. It also demonstrates its robustness against synthetic highly noisy speech datasets.

Study on the Construction of a Time-Space Four-Dimensional Combined Imaging Model and Moving Target Location Prediction Model.

Time-space four-dimensional motion target localization is a fundamental and challenging task in the field of intelligent driving, and an important part of achieving the upgrade in existing target localization technologies. In order to solve the problem of the lack of localization of moving targets in a spatio-temporal four-dimensional environment in the existing spatio-temporal data model, this paper proposes an optical imaging model in the four-dimensional time-space system and a mathematical model of the object-image point mapping relationship in the four-dimensional time-space system based on the central perspective projection model, combined with the one-dimensional “time” and three-dimensional “space”. After adding the temporal dimension, the imaging system parameters are extended. In order to solve the nonlinear mapping problem of complex systems, this paper proposes to construct a time-space four-dimensional object-image mapping relationship model based on a BP artificial neural network and demonstrates the feasibility of the joint time-space four-dimensional imaging model theory. In addition, indoor time-space four-dimensional localization prediction experiments verify the performance of the model in this paper. The maximum relative error rates of the predicted motion depth values, time values, and velocity values of this localization method compared with the real values do not exceed 0.23%, 2.03%, and 1.51%, respectively

Reliability-Based Large-Vocabulary Audio-Visual Speech Recognition.

Audio-visual speech recognition (AVSR) can significantly improve performance over audio-only recognition for small or medium vocabularies. However, current AVSR, whether hybrid or end-to-end (E2E), still does not appear to make optimal use of this secondary information stream as the performance is still clearly diminished in noisy conditions for large-vocabulary systems. We, therefore, propose a new fusion architecture—the decision fusion net (DFN). A broad range of time-variant reliability measures are used as an auxiliary input to improve performance. The DFN is used in both hybrid and E2E models. Our experiments on two large-vocabulary datasets, the Lip Reading Sentences 2 and 3 (LRS2 and LRS3) corpora, show highly significant improvements in performance over previous AVSR systems for large-vocabulary datasets. The hybrid model with the proposed DFN integration component even outperforms oracle dynamic stream-weighting, which is considered to be the theoretical upper bound for conventional dynamic stream-weighting approaches. Compared to the hybrid audio-only model, the proposed DFN achieves a relative word-error-rate reduction of 51% on average, while the E2E-DFN model, with its more competitive audio-only baseline system, achieves a relative word error rate reduction of 43%, both showing the efficacy of our proposed fusion architecture.

Efficient acoustic feature transformation in mismatched environments using a Guided-GAN

We propose a new framework to improve automatic speech recognition (ASR) systems in resource-scarce environments using a generative adversarial network (GAN) operating on acoustic input features. The GAN is used to enhance the features of mismatched data prior to decoding, or can optionally be used to fine-tune the acoustic model. We achieve improvements that are comparable to multi-style training (MTR), but at a lower computational cost. With less than one hour of data, an ASR system trained on good quality data, and evaluated on mismatched audio is improved by between 11.5% and 19.7% relative word error rate (WER). Experiments demonstrate that the framework can be very useful in under-resourced environments where training data and computational resources are limited. The GAN does not require parallel training data, because it utilises a baseline acoustic model to provide an additional loss term that guides the generator to create acoustic features that are better classified by the baseline.

Research on Performance Management Evaluation of Public Sports Venues Based on GFAHP

The importance of public stadiums has become increasingly prominent, but the efficiency of its performance management operation mechanism is low, which affects the overall operation efficiency of the whole stadium. However, due to the diversity and incompleteness of its data types, the traditional analytic hierarchy process is difficult to construct its index system. This article introduces the grey fuzzy analytic hierarchy process. Based on the analysis of its performance management problems, this article constructs the index system and makes an application case analysis. The results show that the GFAHP method (group fuzzy analytic hierarchy process) can ensure the integrity of information data. In data classification, the highest accuracy of sample classification is 40.13%, and the relative error rate is basically less than 6%. The performance effect is much higher than other classification algorithms. At the same time, the total score of the evaluation index system constructed by GFAHP is more than 3, the overall operation is good, and the risk evaluation effect of each influencing index factor is also good. Using the grey analytic hierarchy process to evaluate the operation and performance management of public stadiums and gymnasiums will help managers do a good job in risk control and quality dynamic adjustment, so as to improve the operation efficiency and benefit of stadiums and gymnasiums.

ANALISA DEBIT BANGKITAN MENGGUNAKAN MODEL ARIMA (AUTOREGRESIF INTEGRATED MOVING AVERAGE)

Flow discharge data must be available in a time series and accurate manner, so there should be no empty periods. Therefore, a model is needed that can reconstruct or estimate the flow discharge of the empty period stochastically. One way to solve this problem is by filling in data or data generation.The philosophy of data is to create new data sets based on generally incomplete or short historical data to obtain longer and more complete data. The new complete and long data is made with properties as well as short data as the source (Sri Harto and Sudjarwadi, 1989). Model ARIMA represents three modeling namely of autoregressive model (AR), moving average (MA), and autoregressive and moving average model (ARMA) which has characteristic of two models. First stage modeling ARIMA is testing stationary data, identification model, estimation parameter model and forecasting. Data used in this model Arima is discharge complete station Maribaya DAS Cikapundung Hulu if from years 1978 from the research. Results ARIMA model produces correlation values of 0.657 with a target value of 1. For the absolute relative error rate (KAR) and the average error rate (RMS) each produces a value of 0.0052 and 0.017 with a target value of 0. The ARIMA model can be used to fill in, generate discharge data and can be used to predict future flow rates. In discharge forecasting, the ARIMA model is only able to predict discharge accurately in a short time span. For long-term forecasting, the resulting forecast will tend to be flat (flat/constant).

BPCNN: Bi-Point Input for Convolutional Neural Networks in Speaker Spoofing Detection.

We propose a method, called bi-point input, for convolutional neural networks (CNNs) that handle variable-length input features (e.g., speech utterances). Feeding input features into a CNN in a mini-batch unit requires that all features in each mini-batch have the same shape. A set of variable-length features cannot be directly fed into a CNN because they commonly have different lengths. Feature segmentation is a dominant method for CNNs to handle variable-length features, where each feature is decomposed into fixed-length segments. A CNN receives one segment as an input at one time. However, a CNN can consider only the information of one segment at one time, not the entire feature. This drawback limits the amount of information available at one time and consequently results in suboptimal solutions. Our proposed method alleviates this problem by increasing the amount of information available at one time. With the proposed method, a CNN receives a pair of two segments obtained from a feature as an input at one time. Each of the two segments generally covers different time ranges and therefore has different information. We also propose various combination methods and provide a rough guidance to set a proper segment length without evaluation. We evaluate the proposed method on the spoofing detection tasks using the ASVspoof 2019 database under various conditions. The experimental results reveal that the proposed method reduces the relative equal error rate (EER) by approximately 17.2% and 43.8% on average for the logical access (LA) and physical access (PA) tasks, respectively.

The State of the Science on Artificial Intelligence-Based Dietary Assessment Methods That Use Digital Images: A Scoping Review

ObjectivesHuman error when estimating food intake is a major source of bias in nutrition research. This study provides an overview of literature on the accuracy of artificial intelligence (AI) methods used to analyze digital images of food compared to human coders and ground truth.MethodsThis scoping review included peer-reviewed journal articles reporting AI (e.g., deep learning) and image coding for food analysis. Literature was searched through August 2021 in 4 databases plus reference mining in conference papers and reviews. Eligible articles reported volume, energy, or nutrients estimated from digital food images via fully automatic AI methods. Two investigators independently screened and extracted data. Study characteristics, methods, and type of results were extracted.ResultsIn total, 8,761 unique publications were identified; 35 papers published from 2010 to 2021 (77% after 2015) were included. Preliminary results are reported. Most papers used newly-captured images or created a digital image database (77%). Some used named image databases (e.g., Rakuten18, UEC Food-100); only ImageNet and Inselspital's dataset were used in more than one study. Most (63%) reported relying on a nutrient database to derive nutrient values; 31% specified using USDA FoodData Central databases. The most frequently reported estimation accuracy results were absolute error (AE; 46%), relative error (RE; 40%), correlation coefficient (26%), and error rate (23%). Other reported measures were Bland-Altman plots; actual and estimated values (e.g., volumes, nutrients); and tests of mean differences. Results were reported for calories (AE, 26% of all studies; RE 20%); macronutrients (AE 20%, RE 9%); volume, mass, weight, or area (AE 17%; RE 26%); and salt (AE 6%; RE 6%).ConclusionsSignificant resources have been devoted to investigating the ability of AI methods to conduct accurate dietary assessment using digital food images. Yet, substantial variability in the databases used and results reported prevents quantitative synthesis of overall accuracy across studies. A validated risk of bias tool is needed to compare study quality. Future research should consider using a limited number of valid databases for food images and nutrition information, and reports should at least include absolute and relative error for volume estimations.Funding SourcesSupported by an NIH grant.

Phase response measurement of spatial light modulators based on a Shack-Hartmann wavefront sensor.

It is known that the phase response of spatial light modulators (SLMs) measured by double-beam interferometers is sensitive to mechanical and environmental disturbances. This paper proposes a Shack-Hartmann wavefront sensor (SHWS) method to measure the phase response characteristics of the SLM. The results show that the phase modulation depth measured by the proposed method is 1.7581λ, and 1.7993λ by the Twyman-Green interferometer method. The difference in the phase modulation depth between the two methods is only 0.0412λ, and its relative error rate is 2.29%. It proves that the phase modulation accuracy obtained by the SHWS with lenslets of 73*73 used in this paper is equivalent to that of the Twyman-Green interferometer. Compared with the interferometer method, the SHWS method is simple, compact, and robust, has good real-time performance, and is relatively vibration insensitive. In the future, the SHWS method will play a more important role in the detection of the SLM's phase response.

Linear and nonlinear analyses of heart rate variability signals under mental load

Mental load has an important effect on the efficiency and reliability of human–machine systems. This study discussed in this paper looked at the heart rate variability (HRV) signal changes of subjects under a mental load state, which was used to explore physiological indices of a mental load. An ErgoLAB smart wearable human factor physiological recorder was used to collect the photoplethysmography (PPG) signals of 30 people in a resting state and while implementing the detonation of energetic materials, and HRV signals were extracted from the PPG. First, we used a subjective questionnaire and time perception test to judge the induction of mental load. Then, linear (time-domain and frequency-domain) and nonlinear (Poincaré plot, scatter plot and sample entropy (SampEn)) analysis was performed on the subjects’ HRV signals, and Pearson’s correlation analysis and t-tests were conducted. In a state of mental load, the score of the subjective questionnaire increased significantly (p < 0.01), and the time perception error value (p < 0.01) and the relative error rate (p < 0.05) increased significantly, which proved that the subjects were under mental load. The results show that HR, RRn, SDNN, RMSSD, pNN50, CV, HF, SD1, SD2 and B-- are useful sensitivity parameters to reliably detect whether there is mental load personnel. The research in this paper provides a theoretical basis for the effective identification of mental load. It can also serve as a reference for the study of people’s job reliability under the influence of mental load.

Dynamic modeling and parameter identification for a gantry-type automated fiber placement machine combined with hydraulic balance system and nonlinear friction

A dynamic model considering hydraulic balance system and nonlinear friction is established for a gantry-type automated fiber placement (AFP) machine. To facilitate a comprehensive understanding, the AFP machine is introduced firstly. Due to the vertical movement of large masses, the AFP machine is equipped with a hydraulic balance system, thus a hydraulic model is secondly proposed to reveal the relationship between hydraulic pressure and the movement of the vertical axis. Then, a dynamic equation of the AFP machine is built by the Newton-Euler method, which is subsequently converted into the form of linearized minimum inertia parameters. In addition, the problems of nonlinear friction are also studied, and a segmented friction model is developed. To identify the parameters, issues regarding excitation trajectory, filtering, and identification algorithm are discussed. Corresponding experiments are performed, and the experimental results suggest that the prediction of the dynamic model is highly coincident with measurement data, and their relative error rates are less than 5%, proving the effectiveness of the established model and the identified parameters. Further analysis indicates that the proposed friction model is more in line with the actual situation, and can provide guidance for other friction research.

Multilingual speech recognition for GlobalPhone languages

In this paper, we present our investigations towards the development of multilingual automatic speech recognition (ML ASR) systems using the GlobalPhone database. In addition to GlobalPhone, we have included 4 Ethiopian languages (Amharic, Oromo, Tigrigna and Wolaytta), as well as Uyghur and English in our investigation. In order to see the impact of language relatedness in ML ASR training, we have analyzed both phonetic overlap and morphological complexity of the languages. Deep Neural Network based ML ASR systems have been developed using ML mix, transfer and multitask learning approaches. Relative word error rate (WER) reductions up to 33.21% have been achieved as a result of using resources of other languages in ML acoustic model training. Our experimental results show that languages with small amounts of monolingual training data benefit a lot from ML training. Moreover, using phonetically related languages in ML training is more beneficiary than using phonetically less related languages. It seems that the nature of the corpus (single or mixed domain, noisy or noise free, etc.) has also an impact in ML training although it is not as important as the phonetic relatedness of the languages.

A novel U‐shaped encoder–decoder network with attention mechanism for detection and evaluation of road cracks at pixel level

AbstractAs the most common road distress, cracks have a substantial influence on the integrity of pavement structures. Accurate identification of crack existence and quantification of crack geometry are thus critical for the decision‐making of maintenance measures. This paper proposes a novel neural network for the detection and evaluation of road cracks at pixel level, which combines the advantage of encoding–decoding network and attention mechanism and can thus extract the crack pixels more accurately and efficiently. The proposed network achieves an excellent detection performance with IOU = 92.85%, precision = 96.90%, recall = 95.36%, F1 = 95.53%. Compared with the other advanced networks, the accuracy of the proposed method is substantially enhanced. The quantitative estimation of key geometrical features of cracks including length, width, and area is successfully realized with the development of a prototype of an intelligent mobile system. Compared with the ground truth, the maximum crack width shows the lowest relative error rate, which ranges −31.75%∼28.57%.

Active Correction for Incremental Speaker Diarization of a Collection with Human in the Loop

State of the art diarization systems now achieve decent performance but those performances are often not good enough to deploy them without any human supervision. Additionally, most approaches focus on single audio files while many use cases involving multiple recordings with recurrent speakers require the incremental processing of a collection. In this paper, we propose a framework that solicits a human in the loop to correct the clustering by answering simple questions. After defining the nature of the questions for both single file and collection of files, we propose two algorithms to list those questions and associated stopping criteria that are necessary to limit the work load on the human in the loop. Experiments performed on the ALLIES dataset show that a limited interaction with a human expert can lead to considerable improvement of up to 36.5% relative diarization error rate (DER) for single files and 33.29% for a collection.

The XMUSPEECH System for Accented English Automatic Speech Recognition

In this paper, we present the XMUSPEECH systems for Track 2 of the Interspeech 2020 Accented English Speech Recognition Challenge (AESRC2020). Track 2 is an Automatic Speech Recognition (ASR) task where the non-native English speakers have various accents, which reduces the accuracy of the ASR system. To solve this problem, we experimented with acoustic models and input features. Furthermore, we trained a TDNN-LSTM language model for lattice rescoring to obtain better results. Compared with our baseline system, we achieved relative word error rate (WER) improvements of 40.7% and 35.7% on the development set and evaluation set, respectively.

New Acoustic Features for Synthetic and Replay Spoofing Attack Detection

With the rapid development of intelligent speech technologies, automatic speaker verification (ASV) has become one of the most natural and convenient biometric speaker recognition approaches. However, most state-of-the-art ASV systems are vulnerable to spoofing attack techniques, such as speech synthesis, voice conversion, and replay speech. Due to the symmetry distribution characteristic between the genuine (true) speech and spoof (fake) speech pair, the spoofing attack detection is challenging. Many recent research works have been focusing on the ASV anti-spoofing solutions. This work investigates two types of new acoustic features to improve the performance of spoofing attacks. The first features consist of two cepstral coefficients and one LogSpec feature, which are extracted from the linear prediction (LP) residual signals. The second feature is a harmonic and noise subband ratio feature, which can reflect the interaction movement difference of the vocal tract and glottal airflow of the genuine and spoofing speech. The significance of these new features has been investigated in both the t-stochastic neighborhood embedding space and the binary classification modeling space. Experiments on the ASVspoof 2019 database show that the proposed residual features can achieve from 7% to 51.7% relative equal error rate (EER) reduction on the development and evaluation set over the best single system baseline. Furthermore, more than 31.2% relative EER reduction on both the development and evaluation set shows that the proposed new features contain large information complementary to the source acoustic features.

Dynamic modeling and parameter identification for a gantry-type automated fiber placement machine

Dynamic models play a critical role in the design of model-based controllers, which has a significant effect on the dynamic characteristics of motion equipment. This paper mainly focuses on the dynamic modeling and parameter identification for a gantry-type automated fiber placement (AFP) machine. First, a dynamic modeling process combining prismatic axes and revolute axes is conducted by Newton-Euler method, in which the effects of friction and hydraulic balance system are also considered. Then, as the convenience for parameter identification and the application in linearity control, the methods of dynamic model linearization and determination of minimum inertial parameters based on the multi-body system (MBS) theory are proposed, with which a dynamic model in the form of linearized minimum inertial parameters is consequently established. To identify the parameters in the model, key issues regarding excitation trajectory, filtering, and identification algorithm are discussed in detail. Finally, corresponding experiments are performed on the AFP machine, and experimental results show that there is a good agreement between the prediction of the model and the measurement in actuality. Data analysis shows that except for Z-axis, the relative error rates of the others are not greater than 5%, which proves the effectiveness of the established dynamic model and the identified parameters.

KALIBRASI PARAMETER HIDROLOGI DAERAH ALIRAN SUNGAI BENTUK RADIAL DENGAN APLIKASI HEC-HMS

One of the hydrological responses in a watershed is influenced by the shape of the watershed (DAS). Lempake watershed with an area of 194.50 km2 has a radial shape. So it is necessary to develop a model to determine the factors that influence the watershed with Radial Shapes using HEC HMS. The modeling that has been obtained at once can be used to calculate the inflow that enters the Lempake Dam, especially when there is significant rainfall. This study aims to obtain the value or magnitude of the parameters in the watershed with a radial shape, then the parameters obtained can be used to estimate the design flood discharge in a watershed with a radial shape. The research stage is to calculate the planned flood discharge from rain data from 2010 to 2019 with HEC HMS and then calibrate it with the flood discharge from field observations. from 2010 to 2019. Based on the results of the analysis, it was obtained that the planned flood discharge entering the Lemapke dam was m3/s and from the modeling results, the parameters of the radial type watershed were obtained, namely the CN value of 95 and the Initial Abstraction (Ia) value of 0.11 . This result has a correlation coefficient of 75%. with a relative error rate of 30%.

Optimizing Data Usage for Low-Resource Speech Recognition

Automatic speech recognition has made huge progress recently. However, the current modeling strategy still suffers a large performance degradation when facing the low-resource languages with limited training data. In this paper, we propose a series of methods to optimize the data usage for low-resource speech recognition. Multilingual speech recognition helps a lot in low-resource scenarios. The correlation and similarity between languages are further exploited for multilingual pretraining in our work. We utilize the posterior of the target language extracted from a language classifier to perform data weighing on training samples, which assists the model in being more biased towards the target language during pretraining. Furthermore, dynamic curriculum learning for data allocation and length perturbation for data augmentation are also designed. All these three methods form the new strategy on optimized data usage for low-resource languages. We evaluate the proposed method using rich resource languages for pretraining (PT) and finetuning (FT) the model on the target language with limited data. Experimental results show that the proposed data usage method obtains a 15 to 25&#x0025; relative word error rate reduction for different target languages compared with the commonly adopted multilingual PT+FT method on CommonVoice dataset. The same improvement and conclusion are also observed on Babel dataset with conversational telephone speech, and <inline-formula><tex-math notation="LaTeX">$\sim$</tex-math></inline-formula>40&#x0025; relative character error rate reduction can be obtained for the target low-resource language.

AM Mini-LED Backlight Driving Circuit Using PWM Method With Power-Saving Mechanism

This paper proposes a mini-light-emitting diode (mini-LED) driving circuit that is driven by pulse width modulation (PWM) for the backlights of active-matrix (AM) liquid crystal displays (LCDs). The proposed circuit compensates for threshold voltage (V_TH) variations of low-temperature poly-crystalline silicon thin-film transistors (LTPS TFTs) and the current-resistance (I-R) rise in VSS lines to supply a stable driving current. Operating the mini-LED at the high luminous efficacy by the PWM driving method and setting only the driving TFT on the path of the driving current reduce the power consumption of the backlight. Based on a 2.89-inch LCD panel with an AMLED <inline-formula> <tex-math notation="LaTeX">$48\times48$ </tex-math></inline-formula> backlit module, the TFTs are fabricated, measured, and fitted. Simulation results show that the relative current error rates are all below 4.67&#x0025; when the V_TH of the driving TFT varies by &#x00B1;0.3 V and the VSS rises by 0.5 V. The voltage across VDD and VSS of the proposed circuit is 4.5 V lower than that of the 6T1C compensating driving circuit, so the power consumption of the circuit is at least 27.05&#x0025; lower. Therefore, the proposed driving circuit is well suitable to use in mini-LED backlit LCDs.