A single-channel method for noise reduction on speech is proposed in this paper. A noise added speech is decomposed into frequency signals using the modified DFT. From each frequency signal, signal and noise elements are sequentially detected and then noise subtraction is performed by subtracting the noise element from the signal one. Detectors for the signal and noise elements are implemented by simple averaging circuits. However, the noise level detector needs a coefficient to compensate differences between the detected levels and proper ones. In this paper, we introduce automatic setting of the coefficient into the proposed method. The effectiveness of the proposed noise reduction method is evaluated in computer simulations of speech noise reduction and subjective evaluations using processed signals.

抄録全体を表示

In this paper, an adaptive gait control system for a quadruped robot traversing over irregular terrain with an intermittent crawl pattern using central pattern generator (CPG) networks with motor dynamics is proposed. We develop a new CPG model, in which the inhibitory and excitatory neurons are replaced by two DC motors that drive the joints of the legs directly. The DC motor functions not only as an actuator but also as a component of the CPG model. Four CPG models are mutually connected to each other to control the joints of each leg. Through the adoption of the concept of the center of gravity (COG), the parameters of CPG units can be varied to adjust the period and amplitude of CPG units to realize the intermittent crawl pattern. Using the plural parameters information, CPG units can change the angle range of joints to control the COG in the stable region when the robot walks on irregular terrain without the need for the superfluous calculation of dynamic. Through the experimental results, we confirm that the robot can realize adaptive and stable walking on irregular terrain by using the proposed CPG control system with the signals from the environment.

抄録全体を表示

In this paper we propose a new robust automatic speech recognition (ASR) method using dynamic time warping (DTW) and a nonlinear median filter (NMF). Although conventional DTW is fast and requires no training, its recognition accuracy is limited. The recognition accuracy of conventional DTW algorithms is lower than that of algorithms using the hidden Markov model (HMM) approach under all noisy conditions. Therefore, in order to improve ASR accuracy, in this paper we first employ the short-time energy method to remove nonspeech segments. Then, we deploy a noise-reduction method. Finally, unlike conventional DTW algorithms, which search for the reference word with minimum distance from the unknown speech waveform, we use an NMF and search for the reference word with minimum median distance from the unknown speech waveform. We find that the recognition accuracy of conventional DTW implementations can be improved substantially by the NMF. Our approach yields DTW recognition accuracy similar to that of the HMM techniques in the presence of 10 dB and 20 dB white noise, while there is no complicated training required in the proposed DTW with the NMF.

抄録全体を表示

A method of separating heart sounds, breathing sounds, and bloodstream sounds (intended signals) from the sounds in the human body (biosignals) using microphone sensors is described as a preprocess for detecting circulatory disease such as irregular heart beat (IHB), arterial sclerosis and sleep apnea syndrome (SAS). In this paper, breathing sounds are defined as bronchial sounds. To separate intended signals from biosignals, the independent component analysis (ICA) algorithm and time-frequency masking by the expectation-maximization (EM) algorithm have been used. However, the separation filter in ICA does not work well if the recording environment has considerable reverberation. In addition, time-frequency masking of the EM algorithm is a noise and local solution problem depending on the initial value. Thus, we propose a new algorithm to solve these problems. Experimental results show that our method performs better than ICA and time-frequency masking of the EM algorithm.

抄録全体を表示