TWT: Table with Written Text for Controlled Data-to-Text Generation

Large pre-trained neural models have recently shown remarkable progress in text generation. In this paper, we propose to generate text conditioned on the structured data (table) and a prefix (the written text) by leveraging the pre-trained models. We present a new data-to-text dataset, Table with Written Text (TWT), by repurposing two existing datasets: ToTTo and TabFact. TWT contains both factual and logical statements that are faithful to the structured data, aiming to serve as a useful benchmark for controlled text generation. Compared with existing data-to-text task settings, TWT is more intuitive, the prefix (usually provided by the user) controls the topic of the generated text. Existing methods usually output hallucinated text that is not faithful on TWT. Therefore, we design a novel approach with table-aware attention visibility and copy mechanism over the table. Experimental results show that our approach outperforms state-of-the-art methods under both automatic and human evaluation metrics.

Cepstral Vocal Tract Modelling for Text-To-Speech Synthesis

In this paper we describe a cepstral model of the vocal tract which models both formants and antiformants. The investigated model is more precise compared to the linear prediction model, which models only the formants of the vocal tract. The exponential function is used for the inverse transformation. However, it is difficult to implement this function on a digital signal processor. To solve this issue we use a continued fraction expansion to approximate the exponential function. The transfer function that approximates the exponential function is realized by using the Infinite Impulse Response (IIR) digital filter, in which branches type Finite Impulse Response (FIR) digital filters are included. The coefficients of the FIR digital filters are just the coefficients of the real speech cepstrum. The state-space difference equations are proposed and implemented on a DSP56300 fixed-point digital signal processor (Motorola). Finally, the results of the digital signal processor implementation for chosen vowels and consonants are evaluated.