The authors have previously proposed the normal-incidence absorption coefficient measurement method using multiple microphones which enables a measurement in the frequency range beyond the cut-on frequencies of the higher order acoustic modes in an impedance measurement tube. The maximum measurement frequency of the method is determined by the cut-on frequency of the first radial mode, (0,1) mode, in the cylindrical tube. In this study, in order to extent the measurement frequency range more, the method which can reduce the factor of the (0,1) mode in the sensing signals is investigated. The method consists of two techniques. The first one is a microphone placement way to make the (0,1) mode signal undetectable by placing the acoustic centers of the microphones at a node of the mode. The second one is related to a sound source structure not to excite the (0,1) mode. Using these techniques, the normal-incidence absorption coefficient can be evaluated under the cut-on frequency of the (4,0) mode, which is about 2.89 times higher than that of the conventional two-microphone transfer function method. The results of numerical simulations and experiments are shown to prove the validity and the feasibility of the proposed method.
Previous studies of speech production have approached the acoustic features of fricatives and affricates in the time and spectral domains. To gain an integrated perspective on the consonants, this study examined whether alveolar fricative /s/, alveolar affricate /ʦ/, and alveolo-palatal affricate /ʨ/ can be mapped in a single time-spectral space. The time-domain variables were the rise duration and the sum of steady and decay duration of the consonants. The spectral-domain variable was the mean intensity of the consonants obtained with a one-third-octave bandpass filter with a center frequency of 3,150 Hz. Using these variables, canonical discriminant analysis was performed for /s/, /ʦ/, and /ʨ/ pronounced by a native Japanese speaker. The results showed that /s/, /ʦ/, and /ʨ/ were successfully mapped in the time-spectral space with high discriminant ratio. This means that the Japanese fricative and affricates having different acoustic features can be integrated into a single representation of the time-spectral domain with a good separation.
We have built a singing database that can be used for research purposes. Since recent songs are protected by copyright law, researchers typically use songs that can be used without copyright. With changes to the copyright law in Japan in 2019, we can now release a singing database consisting of songs protected by the law under several restrictions. Our database mainly consists of Japanese pop songs by a professional singer. We collected a total of 50 songs with around 57 minutes of vocals recorded in a studio. After recording, we labeled the phoneme boundaries and converted the songs into the MusicXML format required for the study of statistical parametric singing synthesis. Statistical analysis of the database was then carried out. First, we counted the number of phonemes to clarify their distribution. Second, we performed acoustical analysis on the distribution of pitch, the interval between notes, and duration. Results showed that although the information is biased, the amount of singing is sufficient in light of the findings of a prior study on singing synthesis. The corpus is freely available at our website, https://zunko.jp/kiridev/login.php .