研究者詳細 - 中藤　良久

論文 - 中藤　良久

分割表示161 件中 1 - 20 件目／全件表示 >>

Mobile device-based speech enhancement system using lip-reading 査読有り

Nakahara T., Fukuyama K., Hamada M., Matsui K., Nakatoh Y., Kato Y.O., Rivas A., Corchado J.M.

Advances in Intelligent Systems and Computing 1237 AISC 159 - 167 2021年01月

記述言語：英語掲載種別：研究論文（学術雑誌）

The lip-reading based speech enhancement method for laryngectomees is proposed to improve their communication in an inconspicuous way. First, we developed a simple lip-reading mobile phone application for Japanese using the YOLOv3-Tiny, which can recognize Japanese vowel sequences. Four laryngectomees tested the application, and we confirmed that the system design concept is along the line of the user needs. Second, the user-dependent lip-reading algorithm with very small training data set was developed. Each 36 viseme images were converted into very small data using VAE(Variational Autoencoder), then the training data for the word recognition model was generated. Viseme is a group of phonemes with identical appearances on the lips. Our viseme sequence representation with the VAE was used to be able to adapt users with very small amount of training data set. Word recognition experiment using the VAE encoder and CNN was performed with 20 Japanese words. The experimental result showed 65% recognition accuracy, and 100% including 1 and 2 candidates. The lip-reading type speech enhancement seems appropriate for embedding mobile devices in consideration of both usability and small vocabulary recognition accuracy. st nd

DOI： 10.1007/978-3-030-53036-5_17

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85089724027&origin=inward

Speaker verification method using htm for security system 査読有り

Sakaguchi Y., Hirakawa R., Kawano H., Nakashi K., Nakatoh Y.

Advances in Intelligent Systems and Computing 1253 AISC 160 - 165 2021年01月

担当区分：責任著者記述言語：英語掲載種別：研究論文（学術雑誌）

In recent years, we have been using biometric verification systems in various places such as daily life and businesses. However, it’s insufficient in hospital and food factory to introduction of the security system of the room access control. This is because they wear gloves, masks and hats in hospitals and factories, so they cannot verify faces or fingerprints. In this study, we propose a speaker verification system using HTM (Hierarchical Temporal Memory). We verify speaker and detection of unlearned speakers by experiments. In addition, we consider the use in hospitals and factories, and also evaluate it in a noisy environment. The evaluation values are Precision, Recall, and F-measure. The experimental results show that in the case of 10 learners, F-measure exceeded 0.6 in the classification of 2 classes of learners and non-learners.

DOI： 10.1007/978-3-030-55307-4_25

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85089626810&origin=inward

Effectiveness of dynamic image features for estimating daily conversation atmosphere 査読有り

Natsumi K., Wakita Y., Nakatoh Y.

IEEE International Conference on Artificial Intelligence in Engineering and Technology, IICAIET 2020 2020年09月

記述言語：英語掲載種別：研究論文（国際会議プロシーディングス）

To achieve a conversation atmosphere estimation method, we confirm the usefulness of optical flow values for conversation atmosphere estimation. We extracted only the response scenes from a database of free conversation between elderly individuals, analyzed two-dimensional optical flow parameters, and calculated the standard deviation values (FX-SD, FY-SD) of all frames for each dimension. These values express the dynamic characteristics of conversation gestures. As a result of comparing the FX-SD values and the FY-SD values of 'smooth conversation scene (not-need help (NNH))' with that of 'not-smooth conversation scene (need help (NH))', the 'NNH scenes' values are smaller than those of the 'NH scenes'. We also confirmed the identification rate using support vector machine (SVM). The average identification rate is 72.3%. These results suggest that optical flow parameters are useful for identification, but the performance is not sufficient. In the future, we should evaluate more conversation databases and provide more in-depth discussion of the effectiveness.

DOI： 10.1109/IICAIET49801.2020.9257834

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85098051857&origin=inward

Speech Enhancement System Using Lip-reading 査読有り

Matsui K., Fukuyama K., Nakatoh Y., Kato Y.O.

IEEE International Conference on Artificial Intelligence in Engineering and Technology, IICAIET 2020 2020年09月

記述言語：英語掲載種別：研究論文（国際会議プロシーディングス）

We have been developing a practical speech enhancement system that supports for laryngectomee. By interviewing users we captured essential issues, such as 'utilization of existing device', 'the appearance needs to be inconspicuous', and 'the device should be easy to use'. Considering those user's needs, we plan to use smart phone platform and develop speech enhancement application so that the users are just ordinary looking, and there is no need to buy any additional device. In order to realize such system, the key concept of our proposed system performs lip-reading and speech synthesis. In this study, we examined a lip-reading method that can recognize by registering the words that you want to speak and that is optimized for the user using a small amount of data. 36 viseme images were converted into very small data using VAE(Variational Auto Encoder), then the training data for word recognition model was generated. Viseme is a group of phonemes with identical appearance on the lips. Our viseme sequence representation with VAE was used to be able to adapt users with very small amount of training data set. Word recognition experiment using VAE encoder and CNN was performed with 20 Japanese words. The experimental result showed 65% recognition accuracy, and 100% including 1st and 2nd candidates. The lip-reading type speech enhancement seems appropriate for embedding mobile devices in consideration of both usability and small vocabulary recognition accuracy.

DOI： 10.1109/IICAIET49801.2020.9257839

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85098077233&origin=inward

Programming assistant system using improved flowchart for the visually impaired 査読有り

Hirakawa R., Hirakawa R., Kawano H., Nakashi K., Nakatoh Y.

Advances in Intelligent Systems and Computing 1018 496 - 501 2020年01月

担当区分：責任著者記述言語：英語掲載種別：研究論文（学術雑誌）

© 2020, Springer Nature Switzerland AG. The purpose of this study is to assist the visually impaired to develop applications. There is a method for the visually impaired to develop applications by reading program codes with screen reader. In this method, the visually impaired can read code on line by line and select a line by knocking the up and down arrow key on the keyboard. After that, they can read the processing and describe an instruction before and after the processing. However, such an operation takes them a long time. Therefore, we suggested the system which let develop applications by editing flowcharts. This system converts program code into flowchart and read blocks in the flowchart. In this system, they can select blocks by knocking the left, right, up and down arrow key. We compared the time for rewriting programs between approach and the proposed system. As a result, the time reduced in all programs.

DOI： 10.1007/978-3-030-25629-6_77

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85070016135&origin=inward

Software Log Anomaly Detection Through One Class Clustering of Transformer Encoder Representation 査読有り

Hirakawa R., Tominaga K., Nakatoh Y.

Communications in Computer and Information Science 1224 CCIS 655 - 661 2020年01月

担当区分：責任著者記述言語：英語掲載種別：研究論文（学術雑誌）

© 2020, Springer Nature Switzerland AG. For smart devices such as smartphones and tablets, developing new software using open source software (OSS) is becoming mainstream. While OSS-based development can greatly increase project productivity, it is more difficult to identify the cause of software defects. In this paper, we propose a deep learning model that performs unsupervised learning based on the log data accumulated in the project and calculates the degree of abnormality per line for newly given logs. The proposed method is evaluated using open supercomputer system log data, Blue Gene/L, and the accuracy of the proposed method is compared with the conventional log anomaly detection method using LSTM AutoEncoder. As a result of the comparative experiment, it was found that the proposed method performed better than the conventional method in the two scores of AUROC and F1 Score at the cutoff point.

DOI： 10.1007/978-3-030-50726-8_85

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85088749659&origin=inward

Study on Software Log Anomaly Detection System with Unsupervised Learning Algorithm 査読有り

Hirakawa R., Tominaga K., Nakatoh Y.

Advances in Intelligent Systems and Computing 1152 AISC 122 - 128 2020年01月

担当区分：責任著者記述言語：英語掲載種別：研究論文（学術雑誌）

© The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature Switzerland AG 2020. In recent years, the existence of open source software (OSS) is indispensable for software development. While developer can benefit from functions of OSS, there is a problem that it is very difficult to locate the cause when problems occur. In this study, we propose a method to calculate anomaly score for each line of log data. In our method, the temporal pattern is learned using Hierarchical Temporal Memory, which is an unsupervised real-time learning algorithm, and the anomaly score is obtained based on the internal state of the model. In the experiment, we compare the learning situation in the following three input formats, word ID, word embedding, and sentence embedding. In the experiments using actual log data, it was found that the method with word ID has the highest f1 score and runtime performance, but the precision needs to be improved in order to suppress useless information.

DOI： 10.1007/978-3-030-44267-5_18

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85085196359&origin=inward

Discussion on differences in frequency band required for horizontal sound localization by age 査読有り国際誌

Okuno T., Hirakawa R., Kawano H., Nakashi K., Nakatoh Y.

ACM International Conference Proceeding Series 2019年05月

担当区分：責任著者記述言語：英語掲載種別：研究論文（学術雑誌）

© 2019 Association for Computing Machinery. In this paper, the horizontal sound image experiment of the real sound source was performed under the head resting condition using the band - limited noise using the low pass filter and the band pass filter. Then, the effect of each frequency band on direction perception of the hearing person and the elderly person was clarified. Five people with healthy men in their twenties and five elderly people in their 70s male hearing white noise pass through low-pass filters (2, 3, 4, 5, 6 kHz) and band-pass filters (1-4, 2-4, 2-5 [kHz]) passed through the directional perception experiment. As a result, similarly to the previous study [1], it is difficult to perceive directions sufficiently if two or more frequency bands are included, and it is difficult to perceive the direction of band-pass noise through a band-pass filter of a narrow frequency band. As a result of this research, Direction perception was found that the frequency band around 4 [kHz] and 6 [kHz] is small as an element of direction perception and the frequency band around 5 [kHz] is an important clue for direction perception.

DOI： 10.1145/3325291.3325395

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85075439420&origin=inward

Study on watching system for door-to-door sales detection and risk determination 査読有り国際誌

Hirakawa R., Kawano H., Nakashi K., Nakatoh Y.

ACM International Conference Proceeding Series 2019年05月

担当区分：責任著者記述言語：英語掲載種別：研究論文（学術雑誌）

© 2019 Association for Computing Machinery. In recent years, malicious visit sales are increasing in Japan, targeting elderly people living alone and dementia patients. Since it is difficult for them to deal with problems with their own power, it is necessary for surrounding people to support damage deterrence. So, we aim to realize a watching system that anyone can use if there is only a smart speaker. In this study, we propose a method to detect door-to-door sales from conversation information of elderly people and visitors with the aim of realizing a system to prevent elderly consumers' troubles. In the evaluation experiment, we evaluate how accurately door-to-door sales can be detected out of the conversation dataset. In addition, we try to determine the risk degree by learning the door-to-door sales sample data given the risk label.

DOI： 10.1145/3325291.3325391

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85075470875&origin=inward

Study on method of veering correction by vibration transmission in walking support for visually impaired people 査読有り国際誌

Hirayama T., Hirakawa R., Kawano H., Nakashi K., Nakatoh Y.

ACM International Conference Proceeding Series 2019年05月

担当区分：責任著者記述言語：英語掲載種別：研究論文（学術雑誌）

© 2019 Association for Computing Machinery. In Japan, accidents are occurring when visually impaired people go out. For example, a collision with an obstacle such as a person or a wall, or a fall from the station platform. A current measure for this problem is a voice guidance. However, visually impaired people place importance on auditory information. Therefore, it is difficult for them to hear surrounding sounds during listening the voice guidance. There is a previous research that trying to guide the visually impaired people to their destination by vibration transmission. However, although this study have realized the guidance of the visually impaired people to their destination without interrupting environmental sounds, this method cannot correct the veering in real time. For above reason, we propose a veering correction system by vibration transmission, which does not disturb acquisition of auditory information. This is a system that gives them vibration by wearing a vibration motor on their shoulders and corrects the veering. In this article, we conducted verification experiment of the effectiveness of the veering correction by vibration transmission as a preliminary study in the development of the veering correction system. As a result of the research, we were able to keep the deviations from over 100 cm to less than 100 cm. As a conclusion, vibration transmission is effective as a method of correcting the veering of visually impaired people.

DOI： 10.1145/3325291.3325392

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85075481449&origin=inward

Study on evaluating risks of routes damages at Earthquakes for evacuation guidance system 査読有り国際誌

Mihara Y., Hirakawa R., Kawano H., Nakashi K., Nakatoh Y.

ACM International Conference Proceeding Series 2019年05月

担当区分：責任著者記述言語：英語掲載種別：研究論文（学術雑誌）

© 2019 Association for Computing Machinery. In recent years, the damage caused by the earthquake has become very large in Japan. Secondary disasters such as tsunami and fire are the main causes of death due to the earthquake, both of which are causes of evacuation delay. Rapid evacuation is important in the event of a disaster. However, due to the earthquake, various obstacles will occur in the aisle. Obstacles may occur due to the collapse of houses due to the earthquake and it may take time to pass. If you can obtain information about the aisle in advance and avoid evacuation routes in advance, you can complete evacuation safer and more quickly. As a method of acquiring traffic information, We thought that the use of SNS is useful from the viewpoint of the real-time nature of information and the penetration rate of young people aggressively actively communicating information. In this research, we proposed an evacuation guidance system to evacuate users with a safe shortest route. Then we examined the danger of passing through obstacles. As a result, "road shredder" is evaluated as the most dangerous, and "road crack" is evaluated as the most secure.

DOI： 10.1145/3325291.3325390

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85075445253&origin=inward

In load collapse prevention system Examination of the relationship between truck vibration and load inclination 査読有り国際誌

Inoshita C., Hirakawa R., Kawano H., Nakashi K., Nakatoh Y.

ACM International Conference Proceeding Series 2019年05月

担当区分：責任著者記述言語：英語掲載種別：研究論文（学術雑誌）

© 2019 Association for Computing Machinery. ACM In recent years, the demand for the logistics industry has increased enormously due to the spread of Internet mail order. On the other hand, the number of fatal accidents caused by trucks is more than 10,000 a year in Japan. Above all, there is a problem that there is an annual fatal and fatal accident due to truck load collapse. In this study, using the drive recorder, we calculated the inclination of the package photographed during driving by using KCF of object tracking method of image processing. Comparing the change of the Luggage tilt and the amount of change in 3-axis acceleration attached to the vehicle/ time of each frame in all the data, frames at each maximum value point almost agreed at the time of” When riding over convex part of inclined road”. On the other hand, “When a sudden steering wheel during driving”, a point where the absolute value of the change in package inclination during one frame reaches 0.005 ° and the point where the maximum change occurs after that. It almost coincided with the point where the curve was reached in the video. As for the 3-axis acceleration sensor, there is no difference in tendency between the curve and the straight running, and it is considered as the main cause that the influence of the roughness of the road is greatly influenced rather than the influence at the time of the curve.

DOI： 10.1145/3325291.3325393

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85075447390&origin=inward

Examination of acoustic features for discriminating between real and loudspeaker speeches 査読有り国際誌

Kinoshita Y., Hirakawa R., Kawano H., Nakashi K., Nakatoh Y.

ACM International Conference Proceeding Series 2019年05月

担当区分：責任著者記述言語：英語掲載種別：研究論文（学術雑誌）

© 2019 Association for Computing Machinery. In the train while driving, there are many noises such as noises from the vibration of the vehicle and the shock noise generated at the vehicle connection, and the roar of human beings, etc. In addition to stressing the passengers, these noises it prevents you from listening to the announcements in the train. In this paper, a new noise canceling earphone system aimed at improving in-car announcement speech listening and reducing stress that the noise gives to passengers by separating surrounding speech and in-train announcement speech regarded as noise I devised. In this study, as a preliminary step to complete the system, in order to find the difference between the ambient speech and the announcement speech such as the roar of the passenger, both speeches were compared using acoustic features. As a result, among the acoustic features used in the comparison, differences were found in the spectral centroid, the spectral roll-off, and the intensity and spectral contrast when divided into sub-bands.

DOI： 10.1145/3325291.3325394

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85075462372&origin=inward

Enhancement of electrolarynx speech based on WaveRNN 査読有り国際誌

Urabe E., Hirakawa R., Kawano H., Nakashi K., Nakatoh Y.

ACM International Conference Proceeding Series 2019年05月

担当区分：責任著者記述言語：英語掲載種別：研究論文（学術雑誌）

© 2019 Association for Computing Machinery.ACM In Japan, there are about 30,000 people whose larynxes have been removed. These people are not able to make voice, so they have problems in communication. For example, they cannot communicate their will enough with communication by writing or gestures alone, thereby avoiding communication. However, they can move the mouth and tongue normally. Therefore, they speak using alternative vocalizations. Electrical artificial larynx (EL) is one of the alternative vocalizations. When the vibrator is pressed against the lower part of the jaw, the vibration sound is transmitted in the sound path. This speech method is easy to learn and is cleaner than other speech methods, but this voice sound is bad quality because it has no intonation and it is noisy. In this study, we tried to improve the EL sound quality by using WaveRNN that is one of Deep learning. After making WaveRNN learn human voice, we converted the EL voice. We evaluated the converted voice in a subjective evaluation experiment in 10 words that selected each consonant and vowels in a balanced manner. The evaluation items are: 1. Clarity of consonants, 2. Clarity of vowels, 3. Clarity of whole voice, 4. Natural, 5. Easy to hear. As a result, Clarity and naturalness of consonants improved in words containing two or more unvoiced consonants. On the other hand, in the voiced consonants and vowels, no improvement was seen by the proposed method. Therefore, we think that voice quality was improved in unvoiced consonants. On the other hand, there was no tendency for improvement in voiced consonants and vowels.

DOI： 10.1145/3325291.3325396

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85075454294&origin=inward

Preliminary study of mobile device-based speech enhancement system using lip-reading 査読有り国際誌

Matsunaga Y., Matsui K., Nakatoh Y., Kato Y., Lopez-Sanchez D., Rodriguez S., Corchado J.

Advances in Intelligent Systems and Computing 800 308 - 315 2019年01月

担当区分：責任著者記述言語：英語掲載種別：研究論文（学術雑誌）

© Springer International Publishing AG, part of Springer Nature 2019. Inconspicuous speech enhancement system for laryngectomies using lip-reading is proposed to improve the usability and the speech quality. The proposed system uses a tiny camera on mobile phone and recognize the vowel sequences using lip-reading function. Three types of Japanese vowel recognition algorithms using MLP, CNN, and MobileNets, were investigated. 3,000 image datasets for training and testing were prepared from five persons while uttering discrete vowels. Our preliminary experimental result shows that the MobileNets is appropriate for embedding mobile devices in consideration of a performance both recognition accuracy and calculation cost.

DOI： 10.1007/978-3-319-94649-8_37

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85049977480&origin=inward

Estimating Age-Dependent Degradation Using Nonverbal Feature Analysis of Daily Conversation 査読有り国際誌

Kana N., Wakita Y., Nakatoh Y.

Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) 11582 LNCS 222 - 231 2019年01月

担当区分：責任著者記述言語：英語掲載種別：研究論文（学術雑誌）

© 2019, Springer Nature Switzerland AG. In this paper, we study a system that estimates the degree of decline in the driving ability of elderly people using non-verbal information from daily conversations. It is necessary for us to find the factors that would affect the calculation of the degree of decline that has reached a problematic level for functioning daily life. We focus on the cases where elderly people cannot understand their partner’s speech as their hearing and concentration abilities decrease with age. We analyze the relationship between the degree of understanding of the partner’s speech and the non-verbal characteristic of the response scene. Based on the results of the acoustic analysis of each utterance, the fundamental frequency (F0) and acoustic power levels of when a person can understand their partner’s speech tend to be higher than those when they cannot. The analysis of the synchronization of the head motions shows that brightness value of difference image when a person can understand their partner’s speech is also higher than when they cannot. These results indicate that these non-verbal factors are effective in estimating the decline in the hearing and concentration abilities of the elderly.

DOI： 10.1007/978-3-030-22219-2_17

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85069857489&origin=inward

Potential for Estimating Age-Dependent Degradation using the Acoustic Characteristics of Daily Conversation 査読有り国際誌

Wakita Y., Kondo S., Nakatoh Y.

2018 IEEE 7th Global Conference on Consumer Electronics, GCCE 2018 760 - 763 2018年12月

記述言語：英語掲載種別：研究論文（学術雑誌）

© 2018 IEEE. Our aim is to create a system to estimate the degree of age-dependent degradation in elderly persons based on their daily conversations. We have already confirmed that the time-dependent features of the fundamental frequency (F0) are effective for estimating conversation smoothness. Here we investigated whether these features are also effective for estimating age-dependent degradation. By analyzing free conversational speech data, we confirmed that the standard deviation value of F0 for each utterance is correlated to the average value of F0 when the impression of the speech is regarded as under 75. This correlation was weaker in older individuals. Our results indicate that the above features of F0 would be effective for estimating age-dependent degradation.

DOI： 10.1109/GCCE.2018.8574840

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85060313749&origin=inward

音声スペクトルの時間変化強調による明瞭性改善手法の提案査読有り

川原　竣介、平川　凜、中藤　良久

産業応用工学会論文誌 6 ( 1 ) 51 - 54 2018年03月

Development of hands-free speech enhancement system for both EL-users and esophageal speech users 査読有り国際誌

Matsunaga Y., Matsui K., Nakatoh Y., Kato Y.

Advances in Intelligent Systems and Computing 620 334 - 341 2018年01月

記述言語：英語掲載種別：研究論文（学術雑誌）

© Springer International Publishing AG 2018. A hands-free speech enhancement system for laryngectomies is proposed to improve the usability and the speech quality. The system has a small and right weight transducer, photo-reflector based lip movement level sensors, and a microphone. The sensor outputs are used for generating the transducer turn on/off signal. A simple performance test was conducted if the system can determine the utterance segments accurately, and the result showed good utterance detection performance. Also, we tested with five more subjects and confirmed that the proposed method was well performed even under the speaker-independent condition.

DOI： 10.1007/978-3-319-62410-5_41

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85022188286&origin=inward

Method for Improving Intelligibility of Elderly Speech Based on Formant Frequency Shift 査読有り

TANAKA YUTO, MIZUMACHI MITSUNORI, NAKATOH YOSHIHISA

Journal of the Institute of Industrial Applications Engineers ( The Institute of Industrial Applications Engineers ) 5 ( 1 ) 32 - 36 2017年01月

　前のページ- 次のページ　

1
2
3
4
5
6
7
8
9