The preponderance of matter over antimatter in the early Universe, the dynamics of the supernova bursts that produced the heavy elements necessary for life and whether protons eventually decay -these mysteries at the forefront of particle physics and astrophysics are key to understanding the early evolution of our Universe, its current state and its eventual fate. The Long-Baseline Neutrino Experiment (LBNE) represents an extensively developed plan for a world-class experiment dedicated to addressing these questions.Experiments carried out over the past half century have revealed that neutrinos are found in three states, or flavors, and can transform from one flavor into another. These results indicate that each neutrino flavor state is a mixture of three different nonzero mass states, and to date offer the most compelling evidence for physics beyond the Standard Model. In a single experiment, LBNE will enable a broad exploration of the three-flavor model of neutrino physics with unprecedented detail. Chief among its potential discoveries is that of matter-antimatter asymmetries (through the mechanism of charge-parity violation) in neutrino flavor mixing -a step toward unraveling the mystery of matter generation in the early Universe. Independently, determination of the unknown neutrino mass ordering and precise measurement of neutrino mixing parameters by LBNE may reveal new fundamental symmetries of Nature.Grand Unified Theories, which attempt to describe the unification of the known forces, predict rates for proton decay that cover a range directly accessible with the next generation of large underground detectors such as LBNE's. The experiment's sensitivity to key proton decay channels will offer unique opportunities for the ground-breaking discovery of this phenomenon.Neutrinos emitted in the first few seconds of a core-collapse supernova carry with them the potential for great insight into the evolution of the Universe. LBNE's capability to collect and analyze this high-statistics neutrino signal from a supernova within our galaxy would provide a rare opportunity to peer inside a newly-formed neutron star and potentially witness the birth of a black hole.To achieve its goals, LBNE is conceived around three central components: (1) a new, highintensity neutrino source generated from a megawatt-class proton accelerator at Fermi National Accelerator Laboratory, (2) a fine-grained near neutrino detector installed just downstream of the source, and (3) a massive liquid argon time-projection chamber deployed as a far detector deep underground at the Sanford Underground Research Facility. This facility, located at the site of the former Homestake Mine in Lead, South Dakota, is ∼1,300 km from the neutrino source at Fermilab -a distance (baseline) that delivers optimal sensitivity to neutrino charge-parity symmetry violation and mass ordering effects. This ambitious yet cost-effective design incorporates scalability and flexibility and can accommodate a variety of upgrades and contributions.With its exceptional combi...

No abstract

No abstract

End-to-end (E2E) automatic speech recognition (ASR) models have recently demonstrated superior performance over the traditional hybrid ASR models. Training an E2E ASR model requires a large amount of data which is not only expensive but may also raise dependency on production data. At the same time, synthetic speech generated by the state-of-the-art textto-speech (TTS) engines has advanced to near-human naturalness. In this work, we propose to utilize synthetic speech for ASR training (SynthASR) in applications where data is sparse or hard to get for ASR model training. In addition, we apply continual learning with a novel multi-stage training strategy to address catastrophic forgetting, achieved by a mix of weighted multi-style training, data augmentation, encoder freezing, and parameter regularization. In our experiments conducted on inhouse datasets for a new application of recognizing medication names, training ASR RNN-T models with synthetic audio via the proposed multi-stage training improved the recognition performance on new application by more than 65% relative, without degradation on existing general applications. Our observations show that SynthASR holds great promise in training the state-of-the-art large-scale E2E ASR models for new applications while reducing the costs and dependency on production data.

The diversity of speaker profiles in multi-speaker TTS systems is a crucial aspect of its performance, as it measures how many different speaker profiles TTS systems could possibly synthesize. However, this important aspect is often overlooked when building multi-speaker TTS systems and there is no established framework to evaluate this diversity. The reason behind is that most multi-speaker TTS systems are limited to generate speech signals with the same speaker profiles as its training data. They often use discrete speaker embedding vectors which have a one-to-one correspondence with individual speakers. This correspondence limits TTS systems and hinders their capability of generating unseen speaker profiles that did not appear during training. In this paper, we aim to build multi-speaker TTS systems that have a greater variety of speaker profiles and can generate new synthetic speaker profiles that are different from training data. To this end, we propose to use generative models with a triplet loss and a specific shuffle mechanism. In our experiments, the effectiveness and advantages of the proposed method have been demonstrated in terms of both the distinctiveness and intelligibility of synthesized speech signals.