荒井 隆行

In this study, two perception experiments were conducted to investigate the misperception of Japanese words with devoiced vowels and/or geminate consonants by young and elderly listeners. In Experiment 1, eight young normal-hearing listeners participated under a white-noise condition, and eight elderly listeners participated in Experiment 2. Two types of word sets which consist of combinations of vowels (V = /i, u/) and voiceless consonants (C = /k, t, s/) were used as stimuli. The first word set involved two- or three-mora words and the second word set had 14 minimal pairs of CVC (:) V, where (:) stands for with or without a geminate consonant. The results of both experiments showed that misperception was great for words with devoiced vowels and even greater for words with geminate consonants. In particularly, the misperception of consonants including high frequency components such as /shi/ or /shu/ was observed for elderly listeners.

This study examines the tongue shapes used by Japanese speakers to produce the English liquids/r/ and /l/. Four native Japanese speakers of varying levels of English acquisition and one North American English speaker were recorded both acoustically and with Electromagnetic Articulography. Seven distinct articulation strategies were identified. Results indicate that the least advanced speaker uses a single articulation strategy for both sounds. Intermediate speakers used a wide range of articulations, while the most advanced non-native speaker relied on a single strategy for each sound.

This study tests the perception of geminate consonants for native speakers of Japanese using audio and visual information. A previous study showed that formant transitions associated with the closing gesture of articulators at the end of a preceding vowel are crucial for perception of stop geminate consonants in Japanese. In addition, this study further focuses on visual cues, to test if seeing the closing gesture affects perception of geminate consonants. Based on a perceptual experiment, it is observed that visual information can compensate for a deficiency in geminate consonant auditory information, such as formant transitions. (C) 2017 Acoustical Society of America

Our physical models of the human vocal tract successfully demonstrate theories such as the source-filter theory of speech production, mechanisms such as the relationship between vocal-Tract configuration and vowel quality, and phenomena such as formant frequency estimation. Earlier models took one of two directions: either simplification, showing only a few target themes, or diversification, simulating human articulation more broadly. In this study, we have designed a static, hybrid model. Each model of this type produces one vowel. However, the model also simulates the human articulators more broadly, including the lips, teeth, and tongue. The sagittal block is enclosed with transparent plates so that the inside of the vocal tract is visible from the outside. We also colored the articulators to make them more easily identified. In testing, we confirmed that the vocal-Tract models can produce the target vowel. These models have great potential, with applications not only in acoustics and phonetics education, but also pronunciation training in language learning and speech therapy in the clinical setting.

We have developed two types of mechanical models of the human vocal tract. The first model was designed for the retroflex approximant [r] and the alveolar lateral approximant [l]. It consisted of the main vocal tract and a flapping tongue, where the front half of the tongue can be rotated against the palate. When the tongue is short and rotated approximately 90 degrees, the retroflex approximant [r] is produced. The second model was designed for [b], [m], and [w]. Besides the main vocal tract, this model contains a movable lower lip for lip closure and a nasal cavity with a controllable velopharyngeal port. In the present study, we joined these two mechanical models to form a new model containing the main vocal tract, the flapping tongue, the movable lower lip, and the nasal cavity with the controllable velopharyngeal port. This integrated model now makes it possible to produce consonant sequences. Therefore, we examined the sequence [br], in particular, adjusting the timing of the lip and lingual gestures to produce the best sound. Because the gestures are visually observable from the outside of this model, the timing of the gestures were examined with the use of a high-speed video camera.

People can understand speech under poor conditions, even when successive pieces of the waveform are flipped in time. Using a new method to measure perception of such stimuli, we show that words with sounds based on rapid spectral changes (stop consonants) are much more impaired by reversing speech segments than words with fewer such sounds, and that words are much more resistant to disruption than pseudowords. We then demonstrate that this lexical advantage is more characteristic of some people than others. Participants listened to speech that was degraded in two very different ways, and we measured each person's reliance on lexical support for each task. Listeners who relied on the lexicon for help in perceiving one kind of degraded speech also relied on the lexicon when dealing with a quite different kind of degraded speech. Thus, people differ in their relative reliance on the speech signal versus their pre-existing knowledge. (C) 2016 Elsevier B.V. All rights reserved.

This study investigates how similarly present and absent English phonemes behind noise are perceived by native and non-native speakers. Participants were English native speakers and Japanese native speakers who spoke English as a second language. They listened to English words and non-words in which a phoneme was covered by noise (added; phoneme + noise) or replaced by noise (replaced; noise only). The target phoneme was either a nasal (/m/ and /n/) or a liquid (/l/ and /r/). In experiment, participants listened to a pair of a word (or non-word) with noise (added or replaced) and a word (or non-word) without noise (original) in a row, and evaluated the similarity of the two on an eight-point scale (8: very similar, 1: not similar). The results suggested that both native and non-native speakers perceived the 'added' phoneme more similar to the original sound than the 'replaced' phoneme to the original sound. In addition, both native and non-native speakers restored missing nasals more than missing liquids. In general, a replaced phoneme was better restored in words than non-words by native speakers, but equally restored by non-native speakers. It seems that bottom-up acoustic cues and top-down lexical cues are adopted differently in the phonemic restoration of native and non-native speakers.

As an extension of a series of models we have developed, a mechanical bent vocal-tract model with nasal cavity was proposed for educational and clinical applications, as well as for understanding human speech production. Although our recent studies have focused on flap and approximant sounds, this paper introduced a new model for the consonants [b], [m] and [w]. Because the articulatory gesture of approximants is slow compared to the more rapid movement of plosives, in our [b] and [m] model, the elastic force of a spring is applied to affect the movement of the lower lip block, as was done for flap sounds in our previous studies. The main difference between [b] and [m] is in the velopharyngeal port, which is closed for [b] and open for [m]. In this study, we concluded that 1) a slower manipulation of the lip block is needed for [w], while 2) [b] and [m] require a faster movement, and finally, 3) close-open coordination of the lip and velopharyngeal gestures is important for [m].

'Summer-Holiday Science Square' is a scientific event organized by the National Museum of Nature and Science Japan to cultivate the science literacy of the citizen. Exhibition booths presented by universities, colleges, research societies, and other associations allow visitors to become familiar with science through experiments, handicrafts, and observations. Handicraft workshops are held ten times a day, with groups of up to five people who have picked up numbered tickets. Even the children younger than school-aged are welcome with parent accompaniment. Participants experience various sound resonance phenomena through experiments in 'Let's experience sound resonance!'. In another exhibit, visitors learn about speech production and the visualization of speech through a series of demonstrations. They first learn the principles of vowel production using vocal tract models. Then they see spectrograms of their own speech on-screen using speech analysis software.

The theory of categorical perception of speech sounds traditionally suggests that speech sound discrimination is conducted based on phonemic labeling, which is an abstract speech representation that listeners are hypothesized to have. However, recent research has found that the impact of labeling on perception of an English /-/l/ contrast may depend on surrounding sound contexts: the effects of phonemic labeling may disappear when the speech sounds to be discriminated are presented in a sentence. The purpose of the present research is to investigate (1) the effects of the sound contexts on categorical perception of speech sounds, and (2) cross linguistic extensibility of such an effect. The experiments employed a Japanese voiced stop consonant continuum, i.e., /ba/- /da/, and tested discrimination of sounds on the continuum by native speakers of Japanese. Experiment 2 in particular investigated whether sounds on such a continuum are discriminated in accordance with the labeling when the sound in question is inserted into a sentence. Through experiments, the cross linguistic effects of surrounding sound contexts are found although there may be some exceptional cases. The research proposes reconsideration of the role of labeling mediation in speech perception.

This paper describes vocal-tract models that we have developed and their applications in education in acoustics. First, we grouped the representative models into two major categories depending on their configuration: straight vs. bent. Then, within each category, we discussed the characteristics of each model in terms of its degree of freedom of movement. Subsequently, we review lectures using the vocal-tract models and report the results of tests and questionnaires carried out simultaneously with the lectures, some of which were re-evaluated in this paper. On the basis of the review, we further discuss how education should be carried out using the vocal-tract models, and we made the following conclusions: 1) the models are useful not only for education on sounds themselves but also for phonetic education 2) it is important that appropriate models should be selected depending on specific purposes and 3) it is necessary to continuously develop more models having different properties with wider variations in the future.

高齢者における摩擦音の促音の知覚を調べるために,高齢者と若年者を対象に実験を行い両者の結果を比較した。特に高齢者については最小可聴値と時間分解能を測定し,それぞれが劣化しているか否かによる結果の違いを検討した。実験は二つ行い,一つは摩擦音の長さの弁別実験もう一つはイシ-イッシに対する識別実験を行った。弁別実験の結果より,時間分解能の低下と摩擦音の長さの弁別能に相関が見られた。また,識別実験の結果では高齢者においてイッシをイシに異聴する傾向が見られた。特に,高齢者において最小可聴値の上昇と時間分解能の低下が,識別に影響を及ぼしている可能性が示唆された。

本研究の目的は,促音に先行する母音の出わたりにおける「フォルマント遷移の有無」と「インテンシティの減衰」が促音の知覚に影響を与えるか検証することである。この二つのパラメータ,及び,閉鎖区間を変化させた2音節の無意味語を合成し,聴取実験を行った。その結果,まず,閉鎖区間が長くなると,促音として知覚され易くなることが確認された。次に,フォルマント遷移がない刺激音は,促音として知覚されにくいことが分かり,フォルマント遷移が促音知覚の手がかりになっていることが明らかになった。更に,インテンシティの減衰の緩急は,促音の知覚には大きな影響を与えていないことが示唆された。

隠したい音声(ターゲット)に対し,別の音(マスカ)でマスキングすることによりスピーチプライバシーを維持するサウンドマスキングにおいて,発話者と関係のない第3者による音声が収められたデータベースから音声信号を選択することでマスカを作成する方法を提案した。本手法によって,利用者に自分自身の声をマスカとして使われることによる抵抗感を軽減することが可能となる。提案法に基づきデータベース内の音声に対し,基本周波数をターゲットと類似させ,かつターゲットとのスペクトル距離を大きくさせるなどの処理を施しかマスカに対して,単語了解度試験を行った。結果として,提案したマスカはピンクノイズや無作為に音声を選択したマスカよりもマスキング効率が高くなることが確認できた。

How speech signals are analyzed and represented remains a foundational challenge both for cognitive science and neuroscience. A growing body of research, employing various behavioral and neurobiological experimental techniques, now points to the perceptual relevance of both phoneme-sized (10-40 Hz modulation frequency) and syllable-sized (2-10 Hz modulation frequency) units in speech processing. However, it is not clear how information associated with such different time scales interacts in a manner relevant for speech perception. We report behavioral experiments on speech intelligibility employing a stimulus that allows us to investigate how distinct temporal modulations in speech are treated separately and whether they are combined. We created sentences in which the slow (similar to 4 Hz; S-low) and rapid (similar to 33 Hz; S-high) modulations-corresponding to similar to 250 and similar to 30 ms, the average duration of syllables and certain phonetic properties, respectively were selectively extracted. Although Slow and Shigh have low intelligibility when presented separately, dichotic presentation of Shigh with Slow results in supra-additive performance, suggesting a synergistic relationship between low- and high-modulation frequencies. A second experiment desynchronized presentation of the Slow and Shigh signals. Desynchronizing signals relative to one another had no impact on intelligibility when delays were less than 45 ms. Longer delays resulted in a steep intelligibility decline, providing further evidence of integration or binding of information within restricted temporal windows. Our data suggest that human speech perception uses multi-time resolution processing. Signals are concurrently analyzed on at least two separate time scales, the intermediate representations of these analyses are integrated, and the resulting bound percept has significant consequences for speech intelligibility a view compatible with recent insights from neuroscience implicating multi-timescale auditory processing.

The physical model designed by Umeda and Teranishi simulating an arbitrary shape of the human vocal tract was a straight tube with a set of plastic plates inserted from one side. Although this model has the advantage that users can configure any shape of the vocal tract, manually manipulating several plates simultaneously is difficult. In this study, we present two models extending Umeda and Teranishi's work to overcome this disadvantage. The first model has a straight tube similar to the original Umeda and Teranishi's model, but the weight of the plates enables them to return to resting position automatically. The second model has a bent tube with the oral and pharyngeal cavities connected at 90 degrees. This feature simulates the actual human vocal tract. The plates move back to their original positions by means of spring coils. In both cases, the plates' automatic return movement facilitates manual manipulation as compared to the original Umeda and Teranishi's model.

We applied a physical model of the human vocal tract, which was originally designed for simulating English In, and tested whether the model can produce a certain range of vowels, especially mid front vowels. We first confirmed that the model can produce such vowels with high intelligibility. By changing the tongue height of the model, learners can adjust the vowel quality by listening to the output sounds as well as receiving a tactile sensation. Therefore, we further used the model for the pronunciation training as a hands-on tool for phonetic education based on the consideration that tongue and finger movements are related in terms of motor control. We demonstrated the vowel production using the model and received feedbacks from a group of listeners engaged in phonetic education. The synergetic effect of visual, auditory and tactile sensations was pointed out as an advantage. We then conducted a production experiment, where participants were asked to repeat each vowel they heard and produce that vowel by means of manipulating the vocal-tract model. As results, slight training effects were observed when using the physical model. Specifically, formant frequencies approached target frequencies as the experimental session progressed.

A study was conducted to establish correlation between English proficiency and consonant identification ability during identification of English consonants in intervocalic contexts in multi-speaker babble noise by Japanese listeners. TOEIC scores and the number of months spent living in English-speaking countries were used to measure the Japanese participants' English proficiency. TOEIC was used to measure the Japanese participants? English proficiency, as it was widely used in Japanese universities as a means to evaluate students' English proficiency. The number of months spent in English-speaking countries was also used to investigate the correlation between English consonant identification under the assumption that listeners with a longer experience of residing in English-speaking countries were more used to and in need of perceiving English speech under a more or less adverse listening condition.

A study was conducted to investigate the relationship between broadening of auditory filter bandwidth near 2 kHz and speech discrimination/identification of monosyllables/de/and/ge by young and elderly Japanese listeners. Twenty elderly listeners within an age group ranging from 63 to 80 years and nine young listeners in the age group of 22 to 25 years were used as subjects for the investigations. All participants were native speakers of Japanese who listened to stimuli presented to either the right or left ear with the low pure-tone threshold at 2 kHz. The pure tone thresholds for both ears were measured using an audiometer before the investigations. The auditory filter bandwidth at 2 kHz was also measured using a measurement system of auditory filters.