Korean Tokenization for Beam Search Rescoring in Speech Recognition
This work addresses a specific decoding bottleneck for Korean ASR, offering an incremental improvement in tokenization for language models.
The paper tackles the problem of improving Korean automatic speech recognition (ASR) by proposing a new tokenization method, SkipTC, for language models used in beam-search decoding, which reduces word error rates compared to standard approaches. It also reports the first ASR performance results on a large-scale 7,600-hour Korean speech dataset.
The performance of automatic speech recognition (ASR) models can be greatly improved by proper beam-search decoding with external language model (LM). There has been an increasing interest in Korean speech recognition, but not many studies have been focused on the decoding procedure. In this paper, we propose a Korean tokenization method for neural network-based LM used for Korean ASR. Although the common approach is to use the same tokenization method for external LM as the ASR model, we show that it may not be the best choice for Korean. We propose a new tokenization method that inserts a special token, SkipTC, when there is no trailing consonant in a Korean syllable. By utilizing the proposed SkipTC token, the input sequence for LM becomes very regularly patterned so that the LM can better learn the linguistic characteristics. Our experiments show that the proposed approach achieves a lower word error rate compared to the same LM model without SkipTC. In addition, we are the first to report the ASR performance for the recently introduced large-scale 7,600h Korean speech dataset.