White noise

About: White noise is a research topic. Over the lifetime, 16496 publications have been published within this topic receiving 318633 citations.

WaveGrad: Estimating Gradients for Waveform Generation

Abstract: This paper introduces WaveGrad, a conditional model for waveform generation which estimates gradients of the data density. The model is built on prior work on score matching and diffusion probabilistic models. It starts from a Gaussian white noise signal and iteratively refines the signal via a gradient-based sampler conditioned on the mel-spectrogram. WaveGrad offers a natural way to trade inference speed for sample quality by adjusting the number of refinement steps, and bridges the gap between non-autoregressive and autoregressive models in terms of audio quality. We find that it can generate high fidelity audio samples using as few as six iterations. Experiments reveal WaveGrad to generate high fidelity audio, outperforming adversarial non-autoregressive baselines and matching a strong likelihood-based autoregressive baseline using fewer sequential operations. Audio samples are available at https://wavegrad-iclr2021.github.io/.

The Influence of Interaural Phase Relations upon the Masking of Speech by White Noise

Abstract: If a communication engineer, confronted with a sound wave consisting of speech mixed with audible random noise, were requested to build a device to separate the speech from the noise, he would be hard pressed to produce a mechanism as effective as the human auditory system. But if he were given two waves, one a sample of speech plus a sample of random noise, the other the same speech minus the noise, he would invoke the elementary mathematical (or electronic) processes of addition and subtraction and oblige in short order with noise‐free speech and with speech‐free noise.This paper examines the performance of the (binaural) human auditory system in handling the two‐wave problem. The effectiveness of the solution is judged in terms of the intelligibility of speech heard against a background of white noise. If monaural intelligibility is taken as a standard of comparison, it is found that the advantage of binaural presentation of the speech and the noise depends upon the interaural phase relations. The auditory system handles best the problems that are easiest for the engineer, though not as effectively as the engineer would handle them. Intelligibility is highest with noise plus speech in one ear, noise minus speech (i.e., the noise wave plus the inverted speech wave) in the other. Words are understood almost, but not exactly, as well with speech plus noise in one ear, speech minus noise (i.e., the speech wave plus the inverted noise wave) in the other.These modes of presentation, in which either the speech waves or the noise waves in the two ears are 180 degrees out of phase, yield word articulation scores as much as 25 percentage units higher than the more conventional mode of presentation in which both the speech waves and the noise waves in the two ears are in phase. Observations with other interaural phase relations and with monaural‐binaural presentation of speech and noise are also described.The results suggest a means of providing a small but probably significant improvement in reception whenever speech is heard through earphones in the presence of ambient noise. The scheme is simply to reverse the connections of one of the earphones.The significance of the results for the theory of masking is discussed.

Fast, accurate algorithm for numerical simulation of exponentially correlated colored noise.

Abstract: Traditionally, stochastic differential equations used in the physical sciences have involved Gaussian white noise. ' In recent times, however, white noise has been replaced by colored noise in a variety of contexts. Laser noise problems and first passage time problems have been shown to necessitate the use of colored noise instead of white noise. Even the mathematical foundations for the theory of stochastic differential equations call for colored noise if the Stratonovich perspective is adopted, as it is when physical arguments are invoked. ' In each of these contexts, many speci6c problems require numerical simulation as a component of a complete analysis. This is usually a consequence of nonlinearity and the resulting intractability in purely analytic terms. Consequently, numerical-simulation algorithms have been developed, originally for white noise, and recently for colored noise as well. The simplest type of colored noise to generate is exponentially correlated colored noise. Such noise introduces only one more parameter, the correlation time for the exponential correlation, and it is easily generated by a linear damping equation driven by white noise. Our new algorithm is for this kind of colored noise. In Sec. II we review the white-noise algorithm and the differential version of the exponentially correlated, colored-noise algorithm. In Sec. III we present the integral version of the colored-noise algorithm and demonstrate its superior properties.

Near-field multiple source localization by passive sensor array

Abstract: The localization of multiple near-field sources in a spatially white Gaussian noise environment is studied. A modified two-dimensional (2-D) version of the multiple signal classification (MUSIC) algorithm is used to localize the signal sources; range and bearing. A global-optimum maximum likelihood searching approach to localize these sources is discussed. It is shown that in the single source situation, the covariances of both the 2-D MUSIC estimator and the maximum likelihood estimator (MLE) approach the Cramer-Rao lower bound as the number of snapshots increases to infinity. In the multiple source situation, it is observed that for a high signal-to-noise ratio (SNR) and a large number of snapshots, the root mean square errors (RMSEs) of both localization techniques are relatively small. However, for low SNR and/or small number of snapshots, the performance of the MLE is much superior that of the modified 2-D MUSIC. >