MirasVoice: A bilingual (English-Persian) speech corpus

MirasVoice: A bilingual (English-Farsi) speech corpus Amir Vaheb1, Ali Janalizadeh Choobbasti1, Behnam Sabeti1, S.H.E. Mortazavi Najafabadi1, Saeid Safavi2 1Miras Technologies International NO. 3, 2nd Alley, North Sheikh Bahai St., Tehran, Iran {Amir, Ali, Behnam, Hani}@miras-tech.com, 2 School of Engineering and Technology, University of Hertfordshire College Lane Campus, Hatfield AL10 8PE, Hertfordshire, UK s.safavi@herts.ac.uk Abstract Speech and speaker recognition is one of the most important research and development areas and has received quite a lot of attention in recent years. The desire to produce a natural form of communication between humans and machines can be considered the motivating factor behind such developments. Speech has the potential to influence numerous fields of research and development. In this paper, MirasVoice which is a bilingual (English-Farsi) speech corpus is presented. Over 50 native Iranian speakers who were able to speak in both the Farsi and English languages have volunteered to help create this bilingual corpus. The volunteers read text documents and then had to answer questions spontaneously in both English and Farsi. The text-independent GMM-UBM speaker verification engine was designed in this study for validating and exploring the performance of this corpus. This multilingual speech corpus could be used in a variety of language dependent and independent applications. For example, it can be used to investigate the effects of different languages (Farsi and English) on the performance of speaker verification systems. The authors of this paper have also investigated speaker verification systems performances when using different train/test architectures. Keywords: Speech Processing, Automatic Speaker Verification, Speech Corpus 1. Introduction Automatic speaker verification (ASV) systems architecture offers a flexible and low-cost solution for biometric authen- tication. Although the research and development of ASV systems in recent years have improved their performance to the point of mass-market deployment; due to advances in noise and channel compensation techniques, these systems can be concerned vulnerable to spoofing (Wu et al., 2015). In recent years a number of counter spoofing techniques have been proposed. The reason behind this attention is the development of robust ASV systems for biometric authen- tication which has many applications in the security sector. This problem has been widely studied for English speak- ers, but not for Iranians or users that know multiple lan- guages. There are several audio corpora for dominant lan- guages like English. For instance, the Santa Barbara cor- pus of spoken American English (Bois et al., 2000 2005) which consists of 249,000 words spoken with the transcrip- tions. Another example is the Callhome American English Speech corpus developed by the Linguistic Data Consor- tium (LDC) (Canavan et al., 1997) which is made up of 120 unscripted 30-minute long on the phone conversations that were made in North America. There are also some purely Farsi speech corpora, but none of them are bilingual. For example, (Bijankhan et al., 1994) speech corpus containing recordings of 300 native Farsi speakers from 10 different dialect regions in Iran. In this study, we present MirasVoice which is a bilingual audio corpus in Farsi and English. MirasVoice contains …

Content truncated to 500 words