Searcharxiv⌕ Search

arXiv subjects

Michael Rubinstein

Publications and source records attributed to Michael Rubinstein.

43 records · Page 3Linked to original sources

Looking to Listen at the Cocktail Party: A Speaker-Independent Audio-Visual Model for Speech Separation

We present a joint audio-visual model for isolating a single speech signal from a mixture of sounds such as other speakers and background noise. Solving this task using only audio as input is extremely challenging and does not provide an association of the separated speech signals with speakers in the video. In this paper, we present a deep network-based model that incorporates both visual and auditory signals to solve this task. The visual features are used to "focus" the audio on desired speakers in a scene and to improve the speech separation quality. To train our joint audio-visual model, we introduce AVSpeech, a new dataset comprised of thousands of hours of video segments from the Web. We demonstrate the applicability of our method to classic speech separation tasks, as well as real-world scenarios involving heated interviews, noisy bars, and screaming children, only requiring the user to specify the face of the person in the video whose speech they want to isolate. Our method shows clear advantage over state-of-the-art audio-only speech separation in cases of mixed speech. In addition, our model, which is speaker-independent (trained once, applicable to any speaker), produces better results than recent audio-visual speech separation methods that are speaker-dependent (require training a separate model for each speaker of interest).

cs.SD↗

The highest lowest zero of general L-functions

Stephen D. Miller showed that, assuming the generalized Riemann Hypothesis, every entire $L$-function of real archimedian type has a zero in the interval $\frac12+i t$ with $-t_0 < t < t_0$, where $t_0\approx 14.13$ corresponds to the first zero of the Riemann zeta function. We give an example of a self-dual degree-4 $L$-function whose first positive imaginary zero is at $t_1\approx 14.496$. In particular, Miller's result does not hold for general $L$-functions. We show that all $L$-functions satisfying some additional (conjecturally true) conditions have a zero in the interval $(-t_2,t_2)$ with $t_2\approx 22.661$.

math.NT↗

The distribution of solutions to xy = N mod a with an application to factoring integers

We consider the uniform distribution of solutions $(x,y)$ to $xy=N \mod a$, and obtain a bound on the second moment of the number of solutions in squares of length approximately $a^{1/2}$. We use this to study a new factoring algorithm that factors $N=UV$ provably in $O(N^{1/3+ε})$ time, and discuss the potential for improving the runtime to sub-exponential.

math.NT↗

End-monomer dynamics in semiflexible polymers

Spurred by an experimental controversy in the literature, we investigate the end-monomer dynamics of semiflexible polymers through Brownian hydrodynamic simulations and dynamic mean-field theory. Precise experimental observations over the last few years of end-monomer dynamics in the diffusion of double-stranded DNA have given conflicting results: one study indicated an unexpected Rouse-like scaling of the mean squared displacement (MSD) r^2(t) ~ t^(1/2) at intermediate times, corresponding to fluctuations at length scales larger than the persistence length but smaller than the coil size; another study claimed the more conventional Zimm scaling r^2(t) ~ t^(2/3) in the same time range. We find a novel intermediate dynamical regime where the effective local exponent of the end-monomer MSD, alpha(t) = d log r^2(t)/ d log t, drops below the Zimm value of 2/3 for sufficiently long chains. The deviation from the Zimm prediction increases with chain length, though it does not reach the Rouse limit of 1/2. The qualitative features of this intermediate regime, in particular the variation of alpha(t) with chain and persistence lengths, can be reproduced through a heuristic scaling argument. Anomalously low values of the effective exponent alpha are explained by hydrodynamic effects related to the slow crossover from dynamics on length scales smaller than the persistence length to dynamics on larger length scales.

cond-mat.soft↗

Unexpected scenario of glass transition in polymer globules: an exactly enumerable model

We introduce a lattice model of glass transition in polymer globules. This model exhibits a novel scenario of ergodicity breaking in which the disjoint regions of phase space do not arise uniformly, but as small chambers whose number increases exponentially with polymer density. Chamber sizes obey power law distribution, making phase space similar to a fractal foam. This clearly demonstrates the importance of the phase space geometry and topology in describing any glass-forming system, such as semicompact polymers during protein folding.

cond-mat.dis-nn↗