Creating New Language and Voice Components for the Updated MaryTTS Text-to-Speech Synthesis Platform

Creating New Language and Voice Components for the Updated MaryTTS Text-to-Speech Synthesis Platform Ingmar Steiner, Sébastien Le Maguer Saarland University & DFKI GmbH {steiner, slemaguer}@coli.uni-saarland.de Abstract We present a new workflow to create components for the MaryTTS text-to-speech synthesis platform, which is popular with researchers and developers, extending it to support new languages and custom synthetic voices. This workflow replaces the previous toolkit with an efficient, flexible process that leverages modern build automation and cloud-hosted infrastructure. Moreover, it is compatible with the updated MaryTTS architecture, enabling new features and state-of-the-art paradigms such as synthesis based on deep neural networks (DNNs). Like MaryTTS itself, the new tools are free, open source software (FOSS), and promote the use of open data. Keywords: text-to-speech synthesis (TTS), front-end, multilingual 1. Introduction Over the last 15 years, MaryTTS (Schröder and Trouvain, 2001) has become one of the reference systems for open source text-to-speech synthesis (TTS). Today, it is actively used by researchers working in speech science, human- computer interaction (HCI), and related fields, as well as by professional and enthusiast software developers in free, open source software (FOSS) or enterprise settings. Its popularity is due in part to the number of languages and voices which are freely available as open resources, as well as the possibility of extending it to support new languages and building custom synthetic voices, or even integrating MaryTTS as a component into more complex applications, such as TTS web services, accessibility software, or spoken dialog systems (SDSs). Because of its implementation in the Java programming language, MaryTTS can be used on any device or computer with a Java Runtime Environment (JRE), and its modular design allows developers and users alike to inspect and customize the entire processing pipeline from input text to speech output. However, the number of people who have participated in, and contributed to, MaryTTS development over the years has led to a complex and overburdened system. Conse- quently, a reboot of the system became unavoidable; until now, we focused on restructuring the system core and ex- plained the philosophy behind the new architecture (Le Ma- guer and Steiner, 2017a; Le Maguer and Steiner, 2017b). Independently, the process of creating new synthetic voices and support for new languages in MaryTTS has also fun- damentally evolved since it was presented by Pammi et al. (2010). Therefore, the current paper presents the new lan- guage and voice building workflow for MaryTTS. The remainder of the paper is structured as follows. Sec- tion 2 provides a brief background on build automation in MaryTTS. In Section 3, we present the new workflow to add support for a new language. Then, in Section 4, we focus on the new voice building pipeline. Finally, in Section 5, we present the reorganized source code and project hosting, particularly from a user perspective. 2. Background Development on MaryTTS has adopted several significant paradigms which had become best practice in Java-based software engineering in the years since the project’s incep- tion. These include, dependency management, where required software …

Content truncated to 500 words