Continuous Expressive Speaking Styles Synthesis based on CVSM and MR-HMM

Proceedings of COLING 2016, the 26th International Conference on Computational Linguistics: Technical Papers, pages 369–376, Osaka, Japan, December 11-17 2016. Continuous Expressive Speaking Styles Synthesis based on CVSM and MR-HMM Jaime Lorenzo-Trueba 1,3 Speech Technology Group1 Universidad Politecnica de Madrid Spain jaime@nii.ac.jp Roberto Barra-Chicote 1 Ascension Gallardo-Antolin 2 Signal Theory and Communications2 Universidad Carlos III de Madrid Spain Junichi Yamagishi 3 National Institute of Informatics3 Tokyo Japan Juan M. Montero1 Speech Technology Group1 Universidad Politecnica de Madrid Spain Abstract This paper introduces a continuous system capable of automatically producing the most adequate speaking style to synthesize a desired target text. This is done thanks to a joint modeling of the acoustic and lexical parameters of the speaker models by adapting the CVSM projection of the training texts using MR-HMM techniques. As such, we consider that as long as sufficient variety in the training data is available, we should be able to model a continuous lexical space into a con- tinuous acoustic space. The proposed continuous automatic text to speech system was evaluated by means of a perceptual evaluation in order to compare them with traditional approaches to the task. The system proved to be capable of conveying the correct expressiveness (average adequacy of 3.6) with an expressive strength comparable to oracle traditional expressive speech synthesis (average of 3.6) although with a drop in speech quality mainly due to the semi-continuous nature of the data (average quality of 2.9). This means that the proposed system is capable of improving traditional neutral systems without requiring any additional user interaction. 1 Introduction It is clear that in recent times there has been an increase in the penetration rates of speech technologies and applications based in speech recognition or speech synthesis are more and more common. Speech synthesis systems in particular have improved greatly in terms of speech intelligibility, speech quality or naturalness in neutral read speech situations regardless of the technology (Barra-Chicote, 2011). The problem appears when one needs to develop applications such as dialogue systems or robotic interfaces, for which a more expressive way of speaking is more appropriate. One of the main problems regarding expressive speech synthesis is the vast amount of possibilities that have to be taken into account. Human expressiveness is not a discrete space but a continuous one, and speaking styles vary greatly from person to person and even from time period to time period. As such, obtaining enough data to cover all the requirements can become a very difficult task and scalability a significant problem. This is why statistical parametric speech synthesis is better fitted to the task. While unit-selection based systems have been proven to be more than capable of providing good quality This work is licenced under a Creative Commons Attribution 4.0 International License. License details: http:// creativecommons.org/licenses/by/4.0/ 369 expressive speech (Adell et al., 2012; Andersson et al., 2010), the adaptability of HMM-based systems allows us to better face the scalability problem. The objective of the present paper is to introduce a complete TTS system …

Content truncated to 500 words