Korean Grapheme Unit-based Speech Recognition Using Attention-CTC Ensemble Network

Open-source toolkit for end-to-end Korean speech recognition

“…End-to-end speech recognition is an active area of research. Many research and papers like [7,22] are announced, and multiple ways to improve efficiency and accuracy are proposed. One of them is an attention mechanism, that provides decoder RNN more information when it produces the output tokens [10].…”

“…These datasets are still influential as benchmark datasets these days [34,35,36]. Recently, however, Librispeech [18] is the most popular benchmark speech corpus on which the latest state-of-the-art ASR models are evaluated [22,37,38]. Although they are greatly useful and dependable, existing speech corpora mainly deal with English or non-Korean language.…”

“…Our encoder network is boosted by using deep CNN, which is proposed in [22,23]. The first one called Deep Speech 2 (ds2) extractor, and the second called VGG extractor.…”

“…We hope our research can be a guideline for later works. Also, there was an attempt to perform Korean speech recognition on a grapheme unit basis [22]. To help these researchers, we have configured it to preprocess in character, subword, and grapheme unit.…”

K-Wav2vec 2.0: Automatic Speech Recognition based on Joint Decoding of Graphemes and Syllables

“…This unique writing system allows us to build a Korean ASR model that is based on either graphemes or syllable blocks. So far, most Korean ASR models (Kim et al, 2020;Bang et al, 2020;Ha et al, 2020) were developed with a high-level modeling unit, syllables, and only a few studies attempted to use graphemes (Park et al, 2019b;. According to previous research that investigated modeling units in Korean ASR tasks, syllable-based models outperform grapheme-based models on Zeroth-Korean dataset (51.6 hrs) in most cases (Wang et al, 2020).…”