Search

Single-channel speech separation can be adopted in many applications. Time-frequency (T-F) masking is an effective method for single-channel speech separation. With advancements in deep learning, T-F masks have become used as a training target, achieving notable separation results. Among the numerous masks that have been proposed, the ideal binary mask (IBM), ideal ratio mask (IRM), Wiener filter (WF) and spectral magnitude mask (SMM) are commonly used and have proven effective, though their separation performance varies depending on the speech mixture and separation model. The existing approach mainly utilizes a single network to approximate the mask of the target speech. However, in mixed speech, there are segments where speech is mixed with other speech, segments where speech is mixed with silent intervals, and segments where high signal-to-noise ratio (SNR) speech is mixed due to pauses and variations in the speakers’ intonation and emphasis. In this paper, we attempt to use different networks to handle speech segments containing various mixtures. In addition to the existing network, we introduce a network (using the Rectified Linear Unit as activation functions) to specifically address segments containing a mixture of speech and silence, as well as segments with high SNR speech mixtures. We conducted evaluation experiments on the speech separation of two speakers using the four aforementioned masks as training targets. The performance improvements observed in the evaluation experiments demonstrate the effectiveness of our proposed method based on the joint network compared to the conventional method based on the single network.

Noise reduction methods have been proposed for various loud noises. However, in a quiet indoor environment, even small noises often cause discomfort. One of the small noises that causes discomfort is noise with resonant frequencies. Since resonant frequencies are often high frequencies, it is difficult to apply conventional active noise control methods to them. To solve this problem, we focused on auditory masking, a phenomenon in which synthesized sounds increase the audible threshold. We have performed several studies on reducing discomfort based on auditory masking. However, it was difficult for comfortable sound design to be achieved using the previously proposed methods, even though they were able to reduce feelings of discomfort. Here, we focus on a pleasant sound: music. Comfortable sound design is made possible by introducing music theory into the design of masker signals. In this paper, we therefore propose comfortable sound design based on auditory masking with chord progression and melody generation to match the peak frequencies of dental treatment noises.

An effective approach to addressing the speech separation problem is utilizing a time–frequency (T-F) mask. The ideal binary mask (IBM) and ideal ratio mask (IRM) have long been widely used to separate speech signals. However, the IBM is better at improving speech intelligibility, while the IRM is better at improving speech quality. To leverage their respective strengths and overcome weaknesses, we propose an ideal threshold-based mask (ITM) to combine these two masks. By adjusting two thresholds, these two masks are combined to jointly act on speech separation. We list the impact of using different threshold combinations on speech separation performance under ideal conditions and discuss a reasonable range for fine tuning the thresholds. By using masks as a training target, to evaluate the effectiveness of the proposed method, we conducted supervised speech separation experiments applying a deep neural network (DNN) and long short-term memory (LSTM), the results of which were measured by three objective indicators: the signal-to-distortion ratio (SDR), signal-to-interference ratio (SIR), and signal-to-artifact ratio improvement (SAR). Experimental results show that the proposed mask combines the strengths of the IBM and IRM and implies that the accuracy of speech separation can potentially be further improved by effectively leveraging the advantages of different masks.

With the development of acoustic simulation methods in recent decades, it has become feasible to simulate the sound pressure distribution of loudspeakers before actually setting physical speakers and measuring the sound field. The parametric array loudspeaker (PAL) has attracted attention due to its sharp directivity and unique applications. However, the sound reproduced by PALs is generated by the nonlinear interactions of ultrasound in the air, which makes it difficult to simulate the reproduced sound of a PAL with low computational load. Focusing on the sharp directivity of ultrasound, we extended conventional acoustic ray-tracing methods to consider the self-demodulation phenomenon of PALs. In this study, we developed a visualization method for the demodulated sound of a PAL. Specifically, the demodulated sound pressure distribution can be simulated to estimate and visualize the area covered by the reproduced sound of PAL before setting a real PAL. In the proposed method, acoustic rays were generated sequentially to express the generation of demodulated sound. Therefore, the proposed method is expected to simulate the demodulated sound of a PAL with acceptable accuracy and low calculation complexity. Quantitative evaluation between simulation results and practical measurement has been carried out, and the results demonstrate the effectiveness of the proposed method.

The development of distant-talk measurement systems has been attracting attention since they can be applied to many situations such as security and disaster relief. One such system that uses a device called a laser Doppler vibrometer (LDV) to acquire sound by measuring an object’s vibration caused by the sound source has been proposed. Different from traditional microphones, an LDV can pick up the target sound from a distance even in a noisy environment. However, the acquired sounds are greatly distorted due to the object’s shape and frequency response. Due to the particularity of the degradation of observed speech, conventional methods cannot be effectively applied to LDVs. We propose two speech enhancement methods that are based on two-stage processing with deep neural networks for LDVs. With the first proposed method, the amplitude spectrum of the observed speech is first restored. The phase difference between the observed and clean speech is then estimated using the restored amplitude spectrum. With the other proposed method, the low-frequency components of the observed speech are first restored. The high-frequency components are then estimated by the restored low-frequency components. The evaluation results indicate that they improved the observed speech in sound quality, deterioration degree, and intelligibility.

In this paper, we propose a three-dimensional (3-D) sound image reproduction method based on spherical harmonic (SH) expansion for 22.2 multichannel audio. 22.2 multichannel audio is a 3-D sound field reproduction system that has been developed for ultra-high definition television (UHDTV). This system can reproduce 3-D sound images by simultaneously driving 22 loudspeakers and two sub-woofers. To control the 3-D sound image, vector base amplitude panning (VBAP) is conventionally used. VBAP can control the direction of 3-D sound image by weighting the input signal and emitting it from three loudspeakers. However, VBAP cannot control the distance of the 3-D sound image because it calculates the weight by only considering the image’s direction. To solve this problem, we propose a novel 3-D sound image reconstruction method based on SH expansion. The proposed method can control both the direction and distance of the 3-D sound image by controlling the sound directivity on the basis of spherical harmonics (SHs) and mode matching. The directivity of the 3-D sound image is obtained in the SH domain. In addition, the distance of the 3-D sound image is represented by the mode strength. The signal obtained by the proposed method is then emitted from loudspeakers and the 3-D sound image can be reproduced accurately with consideration of not only the direction but also the distance. A number of experimental results show that the proposed method can control both the direction and distance of 3-D sound images.

The reduction of the aeroacoustic bogie noise that is generated by the Shinkansen high-speed trains travelling faster than 300 km/h is an important environmental issue. Currently, the amount of aeroacoustic bogie noise can be quantitatively estimated at the measuring points based on a spatial distribution of the sound pressure level (SPL) that is obtained by a two-dimensional microphone array in a wind tunnel test. However, it is necessary to appropriately consider noise generation and sound field with respect to various acoustic properties such as ground reflection and insertion loss of a bogie side cover. In this study, the transfer function between the integrated spatial distribution of SPL and the results that are obtained by an omnidirectional microphone are calculated by a numerical method. The aforementioned transfer function is calculated based on the bogie conditions that the side cover would be set and that the ground reflection effect would also be included. The transfer function is observed to change significantly in the lateral direction in low-frequency regions, especially in the 125 and 160 Hz bands. These spatial trends are also confirmed using an acoustic test that is based on the reciprocal theorem, and it is observed that these trends are caused by the interference between the reflection waves from the ground, the cavity upper wall, and the side covers. The SPL of the aeroacoustic bogie noise of Shinkansen trains can be estimated using this transfer function and can be compared with the results that are obtained from the field tests.

A method has been developed for predicting the aerodynamic noise from the bogie of a high-speed train using a two-dimensional microphone array in a low-noise wind tunnel. First, the mean velocity distribution of flow was simulated precisely in the low-noise wind tunnel. Next, aerodynamic noise generated by the bogie, hereinafter referred to as aerodynamic bogie noise, was estimated from the noise source distribution measured with the two-dimensional microphone array. Finally, based on the experimental results, the predicted noise generated from the lower part of the car (i.e. the total of the aerodynamic noise estimated through the proposed method and the rolling and machinery noise estimated in a previous study) was compared with the measurement data obtained near the track in the field test. It was found that the predicted sound pressure level showed good agreement with those measured in the field test. This suggests that the proposed method is appropriate to estimate the aerodynamic bogie noise quantitatively. It was also shown that the contribution of the aerodynamic bogie noise to the total noise generated from the lower part of the car is greater than that of rolling and machinery noise, especially below 500 Hz.