Open-ear headphones that suppress leakage sounds by introducing a passive acoustic circuit and do not interfere with natural sidetones are becoming popular. This type of headphone opens a new experimental scheme that enables the acquisition of produced voiced sounds under various auditory stimulations without capturing the applied stimulations. We conducted objective measurements using manikin and voice acquisition tests with a human participant. We present preliminaly results and discuss prospective applications.

In this study, we investigate the sound source location of the voiceless glottal fricative [h] using aeroacoustic simulations based on 3D vocal tract geometries extracted from MRI. The vocal tract geometry of [h] in /ha/ and /ho/ produced by a Japanese male were analyzed with two glottal configurations: an adducted and a resting state. By solving the compressible Navier-Stokes equations, the airflow and sound generation were simultaneously simulated. Results revealed that primary sound sources appeared from the upper pharynx to the soft palate in /ha/, whereas sound levels decreased by over 20 dB when the glottis opened in /ho/, indicating the glottis is the main sound source in /ho/.

The recent increase in the number of Japanese language learners has placed a greater burden on Japanese language education. Expanding speech education materials can help alleviate the workload of Japanese language teachers. In this study, we evaluated the effectiveness of an authoring system with visual aids that allows learners to engage in selfstudy and recognize issues in their own pronunciation. Using this system, 13 Japanese language learners practiced Japanese sentence intonation. The results showed that while the system had a slight effect in modifying intonation for the acquisition of sentence-final ascent, it was not effective in facilitating the learning of prosodic downstep throughout entire sentences. These findings suggest that the proposed method may be useful for learning localized prosodic elements, such as sentencefinal ascent. Other potential influencing factors include the number of practice sessions and individual differences in phonetic ability, which remain topics for future research.

Fatigue is an inevitable part of daily life, and as of 1999, 60 % of the working population in Japan was aware of fatigue. Furthermore, the use of electronic devices is increasing due to changes in lifestyle, starting with the COVID-19 in 2020, and the recent development of the information society. This is likely to cause additional fatigue in daily life and affect our lives. Therefore, we focus on facial expressions and speech utterances to estimate the degree of fatigue because they’re easily accessible features of the human body. Although fatigue estimation systems using speech recognition and image recognition have been studied and developed separately, those using both recognition results have not been developed yet. In this study, we aimed to develop a system that estimates the degree of fatigue of a user based on the results of each recognition technique, while using machine learning and both speech and image recognition techniques.

Speaker recognition technology can identify the speaker from the voice. In this study, a speech training system using speaker recognition was developed to practice roles. In order to improve the accuracy of speaker recognition, new samples were recorded to evaluate which roles they sound like. The results were used to train a machine learning model. The speaker recognition accuracy of the newly trained model was improved compared with the previous model. Also many subjects responded that they enjoyed the training and they would like to do it again. It was concluded that the new system provides good practice for voice acting. This speech training system allows even beginners to easily practice voice acting.

An AI-based voice counseling system was developed to practice dialogue for people who are not good at conversation. Subjects were asked to dialogue with this system and the voice was recorded. They were also asked to report their current psychological state before and after the dialogue, respectively. Next, the audio recordings of dialogue were analyzed to show the changes in fundamental frequency and intensity during the subjects' dialogue. These results suggest that this dialogue system is effective in promoting pleasant dialogue while easing the tension of the subjects.