A Singing Voice Database in Basque for Statistical Singing Synthesis of Bertsolaritza

A singing voice database in Basque for statistical singing synthesis of bertsolaritza Xabier Sarasola1, Eva Navas1, David Tavarez1, Daniel Erro1,2, Ibon Saratxaga1, Inma Hernaez1 1 AHOLAB, University of the Basque Country (UPV/EHU), Bilbao, Spain 2Basque Foundation for Science (IKERBASQUE), Bilbao, Spain {xabier.sarasola, eva.navas, david.tavarez, daniel.erro, ibon.saratxaga, inma.hernaez}@ehu.eus Abstract This paper describes the characteristics and structure of a Basque singing voice database of bertsolaritza. Bertsolaritza is a popular singing style from Basque Country sung exclusively in Basque that is improvised and a capella. The database is designed to be used in statistical singing voice synthesis for bertsolaritza style. Starting from the recordings and transcriptions of numerous singers, diarization and phoneme alignment experiments have been made to extract the singing voice from the recordings and create phoneme alignments. This labelling processes have been performed applying standard speech processing techniques and the results prove that these techniques can be used in this specific singing style. Keywords: Singing synthesis database, Basque oral resource 1. Introduction Singing voice synthesis has attracted a lot of research and commercial attention in the last years. Two main tech- niques have been applied to the generation of singing voices: unit selection synthesis (Kenmochi and Ohshita, 2007) and hidden Markov model (HMM) based synthesis (Nakamura et al., 2014). Both of them are based on cor- pus and the quality of the produced voice depends heavily on the variety and quality of the corpus. Therefore it is essential to have a suitable database of singing voice that is sufficiently representative to obtain good synthesis re- sults. However, there are no such appropriate databases publicly available for any language. Though a systematic procedure to create a database for expressive singing voice synthesis has been proposed (Umbert et al., 2013), in gen- eral, each group performs its own recordings, such as those used in the works for Japanese (Saino et al., 2006) (Oura et al., 2010), for Spanish (Janer et al., 2006) and for French (d’Alessandro et al., 2014). The final aim of our work is to develop an HMM based singing synthesis system suitable to sing bertsos, a very popular singing style in the Basque Country. There has been a previous first attempt at creating a bertso singing synthesis system in the past (Astigarraga et al., 2013), us- ing a read speech database. Since the quality of the syn- thesized voices strongly depends on the training data, the results achieved by this system are far from satisfactory in terms of naturalness and voice quality. Therefore a new resource has to be developed in order to achieve suitable results. In this paper we present this new database and the results of the first segmentation experiments. The paper is structured as follows: section 2 briefly intro- duces bertsolaritza and the particularities of this style. Sec- tion 3 describes the new singing database. In section 4 the segmentation experiments made on this new database are detailed and, finally, section 5 presents the conclusions and the pending works for the future. 2. Bertsolaritza Bertsolaritza is the …

Content truncated to 500 words