Shunsuke Ishimitsu

Signal processing circuit in an audio device

A signal processing circuit in an audio device includes a variable gain control section for providing a variable amplification gain to an input signal in response to a gain control signal, an input signal level detecting section for detecting a level of the input signal and for generating a corresponding detection signal, and a time constant setting section for storing a plurality of time constant tables. Each one of the plurality of time constant tables includes a set of input signal delay characteristics corresponding to a predetermined level of the input signal. A time constant selecting section selects one of the plurality of time constant tables in the time constant setting section in response to the detection signal. The gain control signal is determined by the selected one of the plurality of time constant tables and the amplification gain of the variable gain control section is varied according to the selected one of the plurality of time constant tables.

Study on the Visualization of the Impression of Button Sounds

A lot of attention has been directed at designing various sounds that are treated as noise, such as automobile acceleration sounds and cleaner sounds. The reason is that the idea of sound being a normal part of product operation has permeated society. We focused on sound design and evaluating it for 11 kinds of button sounds. First, an impression was extracted by the semantic differential (SD) method, and the relevance with a time frequency analysis was investigated. Next, we confirmed whether or not the impression changed when a sound that generated a bad impression was processed using adaptive control into a sound that generated a good impression.

A study of evaluating the button sounds

In recent years, much attention has been directed at the sound design which designs various sound treated as noise, such as automobile acceleration sound and cleaner sound, because the point of view sound is a part of products, has permeated. This research considered the sound design and its evaluation about 11 kinds of the button sounds. First, the impression was extracted by the SD method and relevance with time frequency analysis was investigated. Moreover, we also confirmed that an impression changed, when the sound of a bad impression is processed into the sound of a good impression using adaptive control.

Construction of Speech Support System Using Body-Conducted Speech Recognition for Disorders

The number of disorders associated with cancer of the pharynx has recently increased, and many require surgical removal of the pharynx. Although successful recovery has been reported after surgery, most cases develop speech disabilities. A common solution used to deal with such communication disorders is esophagus vocalization. However, esophagus speech sounds are not a clear type of speech as they do not provide a sufficient volume of speech or frequency for use in daily life. Here, we report the development of a speech support system using body-conducted speech recognition for cases with such a disability. The system uses two features: body-conducted speech recognition and a sub-word unit transfer function database. The system retrieves speech from body-conducted speech. We used a healthy subject to demonstrate the feasibility of making body-conducted speech clear using the sub-word unit transfer function.

TRIANGULAR BIORTHOGONAL WAVELETS WITH EXTENDED LIFTING

We present a new family of triangular biorthogonal wavelets that is defined on a triangular lattice by introducing a new operation to generalize two-dimensional lifting, which we call twist. The resulting filters inherit several remarkable features of the early triangular biorthogonal wavelet filters such as the hexagonal symmetry of low-pass filters, symmetrical arrangement of three high-pass filters on the lattice, and that the wavelet decomposition produces uniform energy distributions over three detail components, preserving the isotropy of decomposed images. Additionally, these filters are a biorthogonal set of truly nonseparable two-dimensional wavelet filters that have much larger support, which provides much larger portions of the total energy to three detail components of decomposed images. We show that this plays a crucial role when extracting the edge structure of an image.

Construction of triangular biorthogonal wavelets using extended lifting

We present new triangular biorthogonal wavelets by extending two-dimensional lifting, which we call twist. The resulting truly two-dimensional biorthogonal filters defined on a triangular lattice inherit several nice features of the early triangular biorthogonal wavelet filters such as the hexagonal symmetry of low-pass filters, symmetrical arrangement of three high-pass filters on the lattice, both of which contribute to preserve isotropy of images in the muniscale wavelet decomposition. Besides, these filters have much larger support, providing much larger portions of the total energy to three detail components of decomposed images. This plays an important role when extracting the edge structure of an image.

Sound quality improvement for the body-conducted speech of a sentence unit using differential acceleration

Speech is a human instrument for communication, and many applications of speech have been proposed. However, the sound quality of speech is reduced by noise in air, and many researchers and engineers have investigated speech measurement methods in noisy environments in terms of signal processing and the use of noise-robust microphones. Conventional approaches of signal measurement can measure speech in an environment with a high signal-to-noise ratio; however, these approaches do not work well in an environment with a low signal-to-noise ratio. By contrast, body-conducted speech, which is speech conducted through the bone and skin of the human body, can be measured in such an environment. However, the frequency characteristics of body-conducted speech are poorer than those of air-conducted speech, and the sound quality needs to be improved for speech applications and conversations. With this background, the paper discusses and investigates sound-quality improvements for the sentence unit of body-conducted speech using differential acceleration, which is a signal retrieval method employed to improve sound quality. The performance of the method is investigated in terms of signal retrieval from body-conducted speech recorded by an accelerometer and Optical Fiber Bragg Grating microphone.

Improvement of Sound Quality on the Body Conducted Speech Using Differential Acceleration

During recent years, applications using speech recognition have been developed to aid dictation during lectures and to advance voice-prompted car navigation systems. Research in speech recognition has been conducted to improve recognition performance and spoken document processing (Nakagawa, 2007). However, even with developments in speech recognition technology, high recognition performance can be compromised due to noisy environments. Standard rate scales, such as CENSREC (Kitaoka et al., 2006) and AURORA (Hirsch and Pearce, 2000), are typically used for evaluating speech recognition performance in noisy environments and have shown that speech recognition rates are approximately 50– 80% when under the influence of noise, demonstrating the difficulty of achieving high recognition percentages. To achieve a high recognition performance, background noise should be minimal, and normal speech should be clear, because the system estimates recognition using a feature vector from its signal. This signal can be affected by sound quality or by an utterance style due to noise in the surrounding environment. When the noise level is low, sound quality becomes clear. However, when the noise level is high, the speech is buried in the noise, causing a change in the speaker’s utterance style, termed the Lombard effect. This change causes the basic frequency to rise because a speaker does not hear the feedback sound from the ear. The method of extracting normal speech under these complex conditions because environment always changes. Several methods have been investigated to extract clear speech under these conditions, such as a noise reduction method, the use of a microphone array or a body-conducted signal. Of these, noise reduction is most commonly used for retrieving a noisy signal and can extract clear speech effectively as long as the background noise is not too high. The microphone array is typically combined with noise reduction. Body-conducted speech is a robust signal extraction method that differs from the other techniques, because it provides a solid signal that propagates through skin and bone. Previously, we built a body-conducted speech recognition system to recognize speech in a noisy environment (98 dB SPL), specifically in the engine room of Oshima-maru, a training ship in Oshima National College of Maritime Technology (Ishimitsu et al., 2004). We found

Body-Conducted Speech Recognition in Speech Support System for Disorders

People with speech disabilities face communication problems in daily conversation. They communicate with substitute speech, which is not used with enough frequency in daily conversation to be readily understood. Therefore, we have proposed the speech support system with bodyconducted speech recognition. The system generates retrieval speech via a transfer function from body-conducted speech when the system uses recognition to decide on a subword sequence and duration time. We needed to examine the effectiveness of body-conducted speech recognition for disorders to construct the system. For the first step in constructing the system, we investigated continuous word unit speech recognition, which used an acoustic model that was not adapted to body-conducted speech for disorders. In this study, we analyzed each parameters of these speeches and experimented with body-conducted speech recognition. We concluded that an adaptation using body-conducted speech recognition to achieve high recognition performance for disorders is valid.

Speech recognition with body‐conducted speech using differential acceleration

Speech‐recognition rates decrease in noisy environments. The body‐conducted speech, conducted in solids such as body and skins, has a noise‐robust characteristics and can be served for recognition systems even in 98 dBSPL‐noise (‐20 dBSNR) environments. However, the body‐conduction could not capture high frequency sounds. Conventional methods for the improvement in sound quality of body‐conducted speeches needs both speeches and body‐conducted speeches. In this paper, a new body‐conducted speech retrieval technique in sound quality without a speech signal itself is proposed. First, high‐frequency components in the body‐conducted speech were emphasized using differential acceleration. Second, conventional noise reduction method was adopted to make a clear body‐conducted speech from a retrieval speech which contains constant noise. The recognition experiments using the proposed method showed that it improved recognition rate in all speakers.