We describe a new stochastic model for generating speech signals suitable for coding at low bit rates. In this model, the speech waveform is represented as a zero mean Gaussian process with slowly-varying power spectr...
详细信息
We describe a new stochastic model for generating speech signals suitable for coding at low bit rates. In this model, the speech waveform is represented as a zero mean Gaussian process with slowly-varying power spectrum. The optimum innovation sequence is obtained by minimizing a subjective error criterion based on properties of human auditory perception. Each block of 40 samples (representing 5 ms of the speech signal sampled at 8 kHz) of the innovation signal is coded into one out of 1024 randomly generated Gaussian sequences of length 40. The chosen sequence minimizes a spectrally weighted error criterion. The innovation signal is thus encoded at 2 kbits/s. A time-varying linear filter whose parameters are determined directly from the speech signal is used to produce the desired power spectrum. Even at this low bit rate the resynthesized speech is barely distinguishable from the original. Wir beschreiben ein neues stochastisches Modell für die Synthese von Sprachsignalen, das für die Codierung bei sehr niedrigen Bitraten geeignet ist. In diesem Modell wird das Sprachsignal durch einen Gauβschen Prozeβ mit langsam veränderlichem Leistungsspektrum dargestellt. Die optimale Innovationsfolge wird durch Minimierung eines subjektiven Fehlerkriteriums erreicht, das auf den Eigenschaften des menschlichen Gehöres beruht. Jeder Block von 40 Abtastwerten des Innovationssignals (das entspricht 5 ms des mit 8 kHz abgetasteten Sprachsignals) wird in eine von insgesamt 1024 verschiedenen zufällig erzeugten Gauβschen Folgen der Länge 40 codiert. Die selektierte Folge minimiert ein spektralgewichtetes Fehlerkriterium. Das Innovationssignal wird also durch 2 kbits/s codiert. Ein zeitvariabler linearer Filter, dessen Parameter unmittelbar aus dem Sprachsignal bestimmt werden, wird benutzt, um das gewünschte Leistungsspektrum zu erzeugen. Selbst bei diesen niedrigen Bitraten ist die synthetische Sprache kaum vom Original zu unterscheiden. Nous décrivons un nouveau modèle stochastiqu
This paper,from the view of a defender,addresses the security problem of cyber-physical systems(CPSs)subject to stealthy false data injection(FDI)attacks that cannot be detected by a residual-based anomaly detector wi...
详细信息
This paper,from the view of a defender,addresses the security problem of cyber-physical systems(CPSs)subject to stealthy false data injection(FDI)attacks that cannot be detected by a residual-based anomaly detector without other defensive *** detect such a class of FDI attacks,a stochastic coding scheme,which codes the sensor measurement with a Gaussian stochastic signal at the sensor side,is proposed to assist an anomaly detector to expose the FDI *** order to ensure the system performance in the normal operational context,a decoder is adopted to decode the coded sensor measurement when received at the controller *** this detection scheme,the residual under the attack can be significantly different from that in the normal situation,and thus trigger an *** design condition of the coding signal covariance is derived to meet the constraints of false alarm rate and attack detection *** minimize the trace of the coding signal covariance,the design problem of the coding signal is converted into a constraint non-convex optimization problem,and an estimation-optimization iteration algorithm is presented to obtain a numerical solution of the coding signal covariance.A numerical example is given to verify the effectiveness of the proposed scheme.
The recording of fired action potentials (spikes) of brain neurons, also known as spike trains, is considered to be the primary mode of information transmission in the nervous system. Electroencephalography (EEG) is t...
详细信息
ISBN:
(数字)9781665459853
ISBN:
(纸本)9781665459853
The recording of fired action potentials (spikes) of brain neurons, also known as spike trains, is considered to be the primary mode of information transmission in the nervous system. Electroencephalography (EEG) is the most direct sampling method for spike trains. However, due to the inherent biological properties of neurons such as spike randomness, timing dynamic, and noisy containment, there are challenges in EEG-related physiological identification tasks (such as sleep staging, epilepsy detection, etc.). Traditional feature engineering of EEG has a tendency toward deterministic statistical analysis and inference, which often ignores the biological properties of neurons. In this paper, we propose an innovative non-deterministic coding method of EEG signals for improving the performance of sleep stage classification tasks. By local normalization, probabilistic sampling, and window projection on the EEG signals, we discretize the continuous signals and feed them into a subsequent classification model. The coding method is tested on the public datasets and typical deep learning models for EEG. Our proposal achieved competitive sleep staging results. The precision of 0.95, 0.84, 0.92, 0.98, and 0.85 were obtained in the Wake, N1, N2, N3, and REM stages, respectively. Our research shows that the non-deterministic coding of EEG has potential for further application in biomedical devices.
Differential Evolution (DE) is a new paradigm of evolutionary algorithm (EA) which has been widely used to solve nonlinear and complex problems. The performance of DE is mainly dependent on the parameter settings, whi...
详细信息
ISBN:
(纸本)9781450311779
Differential Evolution (DE) is a new paradigm of evolutionary algorithm (EA) which has been widely used to solve nonlinear and complex problems. The performance of DE is mainly dependent on the parameter settings, which relate to not only characteristics of the specific problem but also the evolution state of the algorithm. Hence, determining the suitable parameter settings of DE is a promising but challenging task. This paper presents an enhanced algorithm, namely, the stochastic coding differential evolution (SDE), to improve the robustness and efficiency of DE. Instead of encoding each individual as a vector of floating point numbers, the proposed SDE represents each individual by a multivariate normal distribution. In this way, individuals in the population can be more sensible to their surrounding regions and the algorithm can explore the search space region-by-region. In the SDE, a newly designed update operator and a random mutation operator are incorporated to improve the algorithm performance. Traditional DE operators such as the mutation scheme and the crossover operator are also accordingly extended. The proposed SDE has been validated by nine benchmark test functions with different characteristics. Four highly regarded EAs are compared in the experiment study. The comparison results demonstrate the effectiveness and efficiency of the SDE.
Many real-world applications can be modeled as multi-objective optimization problems (MOPs). Applying differential evolution (DE) to MOPs is a promising research topic and has drawn a lot of attention in recent years....
详细信息
ISBN:
(纸本)9781450305570
Many real-world applications can be modeled as multi-objective optimization problems (MOPs). Applying differential evolution (DE) to MOPs is a promising research topic and has drawn a lot of attention in recent years. To search high-quality solutions for MOPs, this paper presents a robust adaptive DE (termed AS-MODE) with following two features. First, a stochastic coding strategy is used to improve the solution quality. This coding strategy represents each individual by a stochastic region, which enables the algorithm to fine-tune solutions efficiently. Second, a probability-based adaptive control strategy is utilized to reduce the influence of parameter settings. The adaptive control strategy associates each parameter with a candidate value set. Better candidate values would have higher selection probabilities to generate new individuals. The performance of the proposed AS-MODE is compared with several highly regarded multi-objective evolutionary algorithms. Simulation results on ten benchmark test functions with different characteristics reveal that AS-MODE yields very promising performance.
The experiment we are presenting here recreates the first part of W.E. cooper''s experiment [6] under the same terms and using the same procedure. It shows to what extent selective adaptation applies to the de...
详细信息
The experiment we are presenting here recreates the first part of W.E. cooper''s experiment [6] under the same terms and using the same procedure. It shows to what extent selective adaptation applies to the detection of place of articulation for French stops. The comparison of our results to W.E. Cooper''s shows that selective adaptation is limited by the nature of the unambiguous stimuli used ad adapters [27, p. 1354]. The shift of the boundary b-d (with adapting stimulus ga) and d-g (with adapting stimulus ba) does exist, but is limited by the acoustic and/or phonetic component. In each of the experiments, the relationship between boundary shifts and the acoustic structure of the stimuli seems to reveal that during the detection of place of articulation, specialized acoustic detectors are present, some specialized in calculating the direction of the transition, and others in determining the magnitude of the slope and/or the location of the locus.
暂无评论