In Other News: A Bi-style Text-to-speech Model for Synthesizing Newscaster Voice with Limited Data

TL;DR

This paper introduces a bi-style neural text-to-speech model capable of synthesizing neutral and newscaster voices with limited data, using style conditioning and contextual embeddings to improve style accuracy.

Contribution

The paper proposes a novel bi-style TTS model that synthesizes multiple speech styles with minimal data, leveraging style conditioning and contextual embeddings.

Findings

Achieves near-natural newscaster-style speech with limited data

Reduces perceived style gap by approximately two-thirds

Outperforms neutral NTTS and concatenative synthesis in style appropriateness

Abstract

Neural text-to-speech synthesis (NTTS) models have shown significant progress in generating high-quality speech, however they require a large quantity of training data. This makes creating models for multiple styles expensive and time-consuming. In this paper different styles of speech are analysed based on prosodic variations, from this a model is proposed to synthesise speech in the style of a newscaster, with just a few hours of supplementary data. We pose the problem of synthesising in a target style using limited data as that of creating a bi-style model that can synthesise both neutral-style and newscaster-style speech via a one-hot vector which factorises the two styles. We also propose conditioning the model on contextual word embeddings, and extensively evaluate it against neutral NTTS, and neutral concatenative-based synthesis. This model closes the gap in perceived style-appropriateness between natural recordings for newscaster-style of speech, and neutral speech synthesis by approximately two-thirds.