A Tool for Making Segmented Speech Corpus for ASR and TTS Modeling

To develop models of Natural Language Processing (NLP), such as speech recognition and speech synthesis, require the provision of a speech corpus that has been segmented and useful as training data. On the other hand, making a speech corpus costs a lot because it can include studio rent and payment of recorded speaker fees. Therefore, in this paper, we develop a tool for making speech corpus equipped with a function that guides the speaker to record their speech in sentences. This tool can be run independently on a personal computer so that we can do recording anytime and anywhere. To produce a better speech corpus, the recording results of this tool still require to be checked because they could potentially have a signal clip or low amplitudes.