Automatic meeting transcription system for the Japanese parliament (diet)

摘要

Applications of automatic speech recognition (ASR) have been extended to a variety of tasks and domains, including spontaneous human-human speech. We have developed an ASR system for the Japanese Pariliament (Diet), which has been operated every day since 2011. By exploiting official records made by human stenographers, we have realized an efficient training scheme of acoustic and language models, which does not require faithful transcripts and thus is scalable to enormous data. This article gives an overview of the system performance and how it changed the meeting record generation process in the Parliament.