A Crowdsourced Open-Source Kazakh Speech Corpus and Initial Speech Recognition Baseline

Abstract

We present an open-source speech corpus for the Kazakh language. The Kazakh speech corpus (KSC) contains around 332 hours of transcribed audio comprising over 153,000 utterances spoken by participants from different regions and age groups, as well as both genders. It was carefully inspected by native Kazakh speakers to ensure high quality. The KSC is the largest publicly available database developed to advance various Kazakh speech and language processing applications. In this paper, we first describe the data collection and preprocessing procedures followed by a description of the database specifications. We also share our experience and challenges faced during the database construction, which might benefit other researchers planning to build a speech corpus for a low-resource language. To demonstrate the reliability of the database, we performed preliminary speech recognition experiments. The experimental results imply that the quality of audio and transcripts is promising (2.8% character error rate and 8.7% word error rate on the test set). To enable experiment reproducibility and ease the corpus usage, we also released an ESPnet recipe for our speech recognition models.

Identifying the influence of transfer learning method in developing an end-to-end automatic speech recognition system with a low data level

“…In [17], a 335 h corpus for the Kazakh language was presented. As a result of the experiment, they showed that a sufficiently large training data set significantly improves the performance of a speech recognition system based on an end-to-end model compared to hybrid ones.…”

A Study of Multilingual End-to-End Speech Recognition for Kazakh, Russian, and English

“…Among the aforementioned three languages, Russian and English are considered resource-rich, i.e., a large number of annotated datasets exist [2,6,31] and extensive studies have been conducted, both in monolingual and multilingual settings [4,25,28]. On the other hand, Kazakh is considered a low-resource language, where annotated datasets and speech processing research have emerged only in recent years [19,26]. The authors of [19] presented the first crowdsourced open-source Kazakh speech corpus and conducted initial Kazakh speech recognition experiments on both DNN-HMM and E2E architectures.…”

“…On the other hand, Kazakh is considered a low-resource language, where annotated datasets and speech processing research have emerged only in recent years [19,26]. The authors of [19] presented the first crowdsourced open-source Kazakh speech corpus and conducted initial Kazakh speech recognition experiments on both DNN-HMM and E2E architectures. Similarly, the authors of [26] presented the first publicly available speech synthesis dataset for Kazakh.…”

“…For Kazakh, we used the recently presented open-source Kazakh Speech Corpus (KSC) [19]. The KSC contains around 332 hours of transcribed audio crowdsourced through the Internet, where volunteers from different regions and age groups were asked to read sentences presented through a web browser.…”

“…In the KSC, all texts are represented using the Cyrillic alphabet, and audio recordings are stored in the WAV format. For the training, validation, and test sets, we used the standard split of non-overlapping speakers provided in [19].…”