1. Jira: a Central Kurdish speech recognition system, designing and building speech corpus and pronunciation lexicon.
- Author
-
Veisi, Hadi, Hosseini, Hawre, MohammadAmini, Mohammad, Fathy, Wirya, and Mahmudi, Aso
- Subjects
AUTOMATIC speech recognition ,SPEECH perception ,SPEECH ,INDO-European languages ,LEXICON ,PRONUNCIATION ,CROWDSOURCING ,OFFICE environment - Abstract
This paper introduces the first large vocabulary speech recognition system (LVSR) for the Central Kurdish language, named Jira. The Kurdish language is an Indo-European language spoken by more than 30 million people in several countries, but due to the lack of speech and text resources, there is no speech recognition system for this language. We introduce the first speech corpus and pronunciation lexicon for the Central Kurdish language to fill this gap. Regarding speech corpus, we designed a sentence collection and the sentences are uttered by 576 speakers in a controlled environment with noise-free microphones [called AsoSoft Speech-Office (A part of this corpus is available for public access at https://github.com/AsoSoft/AsoSoft-Speech-Corpus)] and in Telegram social network environment using mobile phones (denoted as AsoSoft Speech-Crowdsourcing), resulting in 43.68 h of speech. Besides, a test set including 11 different document topics is designed and recorded in two corresponding speech conditions [i.e., Office (The office test set is available online at https://github.com/AsoSoft/AsoSoft-Speech-Testset) and Crowdsourcing]. Furthermore, a 60 K pronunciation lexicon is prepared in this research in which we faced several challenges and proposed solutions for them. Our methods for script standardization of lexical variations and automatic pronunciation of the lexicon tokens are presented in detail. To set up the recognition engine, we used the Kaldi toolkit. A statistical tri-gram language model is used and several standard recipes including HMM-based models (i.e., mono, tri1, tr2, tri2, tri3), SGMM, and DNN methods are used to generate the acoustic model. These methods are trained with AsoSoft Speech-Office and AsoSoft Speech-Crowdsourcing and a combination of them. The SGMM acoustic model achieved the best performance, which results in 13.9% of the average word error rate (on different document topics) and 4.9% for the general topic. [ABSTRACT FROM AUTHOR]
- Published
- 2022
- Full Text
- View/download PDF