This paper introduces a large-scale spontaneous speech corpus of Korean, named KsponSpeech. This corpus contains 969 h of general open-domain dialog utterances, spoken by about 2000 native Korean speakers in a clean environment. All data were constructed by recording the dialogue of two people freely conversing on a variety of topics and manually transcribing the utterances. The transcription provides a dual transcription consisting of orthography and pronunciation, and disfluency tags for spontaneity of speech, such as filler words, repeated words, and word fragments. This paper also presents the baseline performance of an end-to-end speech recognition model trained with KsponSpeech. In addition, we investigated the performance of standard end-to-end architectures and the number of sub-word units suitable for Korean. We investigated issues that should be considered in spontaneous speech recognition in Korean. KsponSpeech is publicly available on an open data hub site of the Korea government.

No abstract

As deep learning-based speech recognition systems are spotlighted, the need for large-scale speech databases for acoustic model training is increasing. Broadcast data can be easily used for database construction, since it contains transcripts for the hearing impaired. However, the subtitle timestamps have not been used to extract speech data because they are often inaccurate due to the inherent characteristics of closed captioning. Thus, we propose to build a large-scale speech database from multi-genre broadcast data with inaccurate subtitle timestamps. The proposed method first extracts the most likely speech intervals by removing subtitle texts with low subtitle quality index, concatenating adjacent subtitle texts into a merged subtitle text, and adding a margin to the timestamp of the merged subtitle text. Next, a speech recognizer is used to extract a hypothesis text of a speech segment corresponding to the merged subtitle text, and then the hypothesis text obtained from the decoder is recursively aligned with the merged subtitle text. Finally, the speech database is constructed by selecting the sub-parts of the merged subtitle text that match the hypothesis text. Our method successfully refines a large amount of broadcast data with inaccurate subtitle timestamps, taking about half of the time compared with the previous methods. Consequently, our method is useful for broadcast data processing, where bulk speech data can be collected every hour.

This paper proposes an automatic method to refine broadcast data collected every week for efficient acoustic model training. For training acoustic models, we use only audio signals, subtitle texts, and subtitle timestamps accompanied by recorded broadcast programs. However, the subtitle timestamps are often inaccurate due to inherent characteristics of closed captioning. In the proposed method, we remove subtitle texts with low subtitle quality index, concatenate adjacent subtitle texts into a merged subtitle text, and correct the timestamp of the merged subtitle text by adding a margin. Then, a speech recognizer is used to obtain a hypothesis text from the speech segment corresponding to the merged subtitle text. Finally, the refined speech segments to be used for acoustic model training, are generated by selecting the subparts of the merged subtitle text that matches the hypothesis text. It is shown that the acoustic models trained by using refined broadcast data give significantly higher speech recognition accuracy than those trained by using raw broadcast data. Consequently, the proposed method can efficiently refine a large amount of broadcast data with inaccurate timestamps taking about half of the time, compared with the previous approaches.

This paper proposes a method to improve speech recognition performance by extracting various pronunciations of the pseudo-morpheme unit from an eojeol unit corpus and generating a new recognition unit considering pronunciation variations. In the proposed method, we first align the pronunciation of the eojeol units and the pseudo-morpheme units, and then expand the pronunciation dictionary by extracting the new pronunciations of the pseudo-morpheme units at the pronunciation of the eojeol units. Then, we propose a new recognition unit that relies on pronunciation by tagging the obtained phoneme symbols according to the pseudo-morpheme units. The proposed units and their extended pronunciations are incorporated into the lexicon and language model of the speech recognizer. Experiments for performance evaluation are performed using the Korean speech recognizer with a trigram language model obtained by a 100 million pseudo-morpheme corpus and an acoustic model trained by a multi-genre broadcast speech data of 445 hours. The proposed method is shown to reduce the word error rate relatively by 13.8% in the news-genre evaluation data and by 4.5% in the total evaluation data.