arXiv subjects

Via Nielson

Publications and source records attributed to Via Nielson.

3 recordsLinked to original sources

arXiv2023

An Experimental Study: Assessing the Combined Framework of WavLM and BEST-RQ for Text-to-Speech Synthesis

We propose a new model architecture specifically suited for text-to-speech (TTS) models. We combine WavLM, a pre-trained self-supervised learning (SSL) speech model, and the BEST-RQ vector quantization framework. We assess the extent to which the more task-agnostic WavLM, coupled with the superior suitability of the simplistic BEST-RQ framework for a wider array of downstream tasks, yields favorable outcomes. Experiments on the LibriSpeech dataset with SUPERB benchmarking assert that the proposed model significantly underperforms. We speculate the underlying reason for this performance is related to the difference between featurizing raw audio waveforms and spectrograms with a quantizer. We discuss the limitations of this approach to better guide future advancements in TTS.

cs.SD↗

arXiv2022

Optical and X-ray GRB Fundamental Planes as Cosmological Distance Indicators

Gamma-Ray Bursts (GRBs), can be employed as standardized candles, extending the distance ladder beyond Supernovae Type Ia (SNe Ia, $z=2.26$). We standardize GRBs using the 3D fundamental plane relation (the Dainotti relation) among the rest-frame end time of the X-ray plateau emission, its corresponding luminosity, and the peak prompt luminosity. Combining SNe Ia and GRBs, we constrain $\Omega_{\text{M}}= 0.299 \pm 0.009$ assuming a flat $\Lambda$CDM cosmology with and without correcting GRBs for selection biases and redshift evolution. Using a 3D optical Dainotti correlation, we find this sample is as efficacious in the determination of $\Omega_{\text{M}}$ as the X-ray sample. We trimmed our GRB samples to achieve tighter planes to simulate additional GRBs. We determined how many GRBs are needed as standalone probes to achieve a comparable precision on $\Omega_{\text{M}}$ to the one obtained by SNe Ia only. We reach the same error measurements derived using SNe Ia in 2011 and 2014 with 142 and 284 simulated optical GRBs, respectively, considering the errorbars on the variables halved. These error limits will be reached in 2038 and in 2047, respectively. Using a doubled sample (obtained by future machine learning approaches allowing a lightcurve reconstruction and the estimates of GRB redhifts when z is unknown) compared to the current sample, with errorbars halved we will reach the same precision as SNe Ia in 2011 and 2014, now and in 2026, respectively. If we consider the current SNe precision, this will be reached with 390 optical GRBs by 2054.

astro-ph.CO↗

arXiv2021

A new perspective on cosmology through Supernovae Ia and Gamma Ray Bursts

The actual knowledge of the structure and future evolution of our universe is based on the use of cosmological models, which can be tested through the so-called 'probes', namely astrophysical phenomena, objects or structures with peculiar properties that can help to discriminate among different cosmological models. Among all the existing probes, of particular importance are the Supernovae Ia (SNe Ia) and the Gamma Ray Bursts (GRBs): the former are considered among the best standard candles so far discovered but suffer from the fact that can be observed until redshift $z=2.26$, while the latter are promising standardizable candles which have been observed up to $z=9.4$, surpassing even the farthest quasar known to date, which is at $z=7.64$. The standard candles can be used to test the cosmological models and to give the expected values of cosmological parameters, in particular the Hubble constant value. The Hubble constant is affected by the so-called \say{Hubble constant tension}, a discrepancy in more than 4 $\sigma$ between its value measured with local probes and its value measured through the cosmological probes. The increase in the number of observed SNe Ia, as well as the future standardization of GRBs through their correlations, will surely be of help in alleviating the Hubble constant tension and in explaining the structure of the universe at higher redshifts. A promising class of GRBs for future standardization is represented by the GRBs associated with Supernovae Ib/c, since these present features similar to the SNe Ia class and obey a tight correlation between their luminosity at the end of the plateau emission in X-rays and the time at the end of the plateau in the rest-frame.

astro-ph.CO↗