Acoustic models based on long short-term memory recurrent neural networks (LSTM-RNNs) were applied to statistical parametric speech synthesis (SPSS) and showed significant improvements in naturalness and latency over those based on hidden Markov models (HMMs). This paper describes further optimizations of LSTM-RNN-based SPSS for deployment on mobile devices; weight quantization, multi-frame inference, and robust inference using an -contaminated Gaussian loss function. Experimental results in subjective listening tests show that these optimizations can make LSTM-RNN-based SPSS comparable to HMM-based SPSS in runtime speed while maintaining naturalness. Evaluations between LSTM-RNNbased SPSS and HMM-driven unit selection speech synthesis are also presented.

No abstract

Abstract. In this paper we study a notion of a κ-covering in connection with Bernstein sets and other types of nonmeasurability. Our results correspond to those obtained by Muthuvel in [7] and Nowik in [8]. We consider also other types of coverings. Definitions and notationIn 1993 Carlson in his paper [3] introduced a notion of κ-coverings and used it for investigating whether some ideals are or are not κ-translatable. Later on κ-coverings were studied by other authors, e.g. Muthuvel (cf. [7]) and Nowik (cf.[8], [9]). In this paper we present new results on κ-coverings in connection with Bernstein sets. We also introduce two natural generalizations of the notion of κ-coverings, namely κ-S-coverings and κ-I-coverings.We use standard set-theoretical notation and terminology from [1]. Recall that the cardinality of the set of all real numbers R is denoted by c. The cardinality of a set A is denoted by |A|. If κ is a cardinal number then

Abstract. In this paper we discuss various questions connected with translations of subsets of the real line. Most of these questions originate from W. Sierpiński. We discuss the number of translations a single subset of the reals may have. Later we discuss almost invariant subsets of Abelian groups.