Research on Two-Dimensionalization Algorithms for Improving Emotion Recognition Accuracy in Speech Data and its Evaluation of Generalized Deployment

Zijun, Yang

WEKO3

lat lon distance

[[sub_check.contents]]

[[sub_radio.contents]]

Field does not validate

[[sub_attr.contents]]　

インデックスツリー

アイテム

Research on Two-Dimensionalization Algorithms for Improving Emotion Recognition Accuracy in Speech Data and its Evaluation of Generalized Deployment

https://doi.org/10.18997/0002000700

名前 / ファイル	ライセンス	アクション
kou_k_589.pdf (16.6 MB)

アイテムタイプ

学位論文 = Thesis or Dissertation(1)

公開日

2024-05-28

資源タイプ

資源タイプ識別子

http://purl.org/coar/resource_type/c_db06

資源タイプ

doctoral thesis

タイトル

Research on Two-Dimensionalization Algorithms for Improving Emotion Recognition Accuracy in Speech Data and its Evaluation of Generalized Deployment

言語

その他のタイトル

感情認識精度向上のための音声データの二次元化アルゴリズムの研究およびその汎用展開の評価

言語

eng

著者

Zijun, Yang

抄録

内容記述タイプ

Abstract

内容記述

With the development of society and the intensification of competition, people face
increasing life pressure in their daily life, which has a significant impact on the mental
health of individuals. This study is dedicated to exploring how this psychosocial health
issue can be attended to and addressed through speech emotion recognition. Speech, as a
natural and intuitive way of expressing emotions, has been found to contain up to 38% of
emotional information. Through in-depth sentiment analysis of speech, we can better understand
the emotional state of individuals and provide feedback accordingly, thus helping
to alleviate the stress they face in life In this study, we innovatively start from speech and
use a novel time series analysis method to transform speech time series into 2D images.
In this process, we employed Hilbert curves to map the time series to the image space. In
this way, we successfully capture the dynamic features of speech into static images, which
lays the foundation for subsequent emotion recognition. In order to realize the accurate
recognition of speech emotion, we designed a neural network suitable for this image representation.
This neural network can effectively extract the key features in the image, thus
realizing the recognition of different emotions. Through a large number of experiments,
we have proved that our method has achieved remarkable results in speech emotion recognition,
providing a solid foundation for further research and application. Not only that,
this study also optimizes other time series imaging algorithms. We improved the Gram’s
Corner Field algorithm by using different downsampling techniques and designed a neural
network model for Gram’s Corner Field. This optimization makes our method more versatile
and able to adapt to different time series data, providing a wider range of possibilities
for future applications. In order to understand the individual’s emotional state more comprehensively,
this study introduces the CyTex method in the extension of the method and
incorporates the concept of speech rate for the segmentation of time series. This innovative
approach further improves the accuracy of speech emotion recognition and lays a
solid foundation for future applications. In the segmentation processing of time series, we
adopt the CyTex method, which effectively divides the time series while maintaining its
continuity. This segmentation allows the neural network to learn the emotional information
in each time period more precisely. Compared with the traditional holistic learning
method, segmentation learning is more capable of capturing the subtle differences of emotional
changes in speech, making the recognition results more accurate. At the same time,
we introduce the concept of speech rate as a new analytical dimension to be incorporated
into the time-series features. Speech rate is not only a surface feature of speech, but it also
combines short-time features and rhythmic features to reflect the emotional information
in speech more comprehensively. By considering speech rate in segmentation learning, we
enable the neural network to be more sensitive to capturing emotional changes in speech, hus improving the accuracy of recognition. This approach experimentally demonstrates
that the segmental learning approach, which introduces CyTex and speech rate, performs
well in the speech emotion recognition task compared to the traditional holistic learning
approach. This provides a more refined and accurate processing means for future speech
emotion recognition applications and lays a more solid foundation for practical applications.
Therefore, by adopting the CyTex method and introducing the concept of speech
rate, we analyze the time series more carefully, which makes our algorithm achieve more
satisfactory results in the emotion recognition task. This innovative approach provides
new perspectives and methods in the field of speech emotion recognition and brings wider
possibilities for future research and applications. This research transcends the confines of
speech emotion recognition, extending its applicability to the realm of brainwave analysis.
The methodologies, initially designed for speech, prove to be versatile as they are successfully
applied to brainwave time series, achieving remarkable results in the identification of
distinct epileptic seizure types. This breakthrough not only signifies the adaptability and
efficacy of the proposed methods but also opens new avenues for applications in neurology
and clinical diagnostics. In achieving excellence in epileptic seizure type recognition,
the study sets the stage for future endeavors aimed at identifying depressive states and
discerning emotional nuances through brainwave analysis. The envisioned expansion of
research activities in these directions reflects the commitment to pushing the boundaries
of knowledge and practical applications in mental health research. This forward momentum
not only enhances our understanding of neurological disorders but also holds promise
for the development of novel diagnostic tools and therapeutic interventions. The exploration
of brainwave signals emerges as a powerful avenue for gaining profound insights into
an individual’s mental state and emotional experiences. Through meticulous analysis of
brainwave patterns, this study provides a nuanced understanding of cognitive processes,
presenting itself as a valuable tool for researchers in psychology and neuroscience. The
nuanced nature of brainwave data offers a rich tapestry of information, shedding light
on the intricate interplay of emotions and mental states. In conclusion, this study, with
a comprehensive scope spanning speech emotion recognition to brainwave analysis, has
reached a pivotal milestone by excelling in epileptic seizure type identification. The transformative methodologies introduced in speech analysis seamlessly extend to the realm of
brainwave time series, opening up new vistas for exploration. The fusion of innovative
approaches with optimized time series imaging algorithms not only enables accurate emotional
state recognition but also propels the research landscape into promising territories
within neurology and mental health. With a commitment to ongoing research, the study
serves as a beacon for future investigations, offering a wealth of tools and insights for
understanding, mitigating, and addressing various aspects of individual life stress, mental
health, and neurological disorders.

内容記述タイプ

TableOfContents

内容記述

1 Introduction| 2 Proposal 1: Speech Emotion Recognition Based on Gramian Angular Field| 3 Proposal 2: Speech Emotion Recognition Based on CyTex and Speech Rate| 4 Proposal 3: Speech Emotion Recognition Based on Hilbert Curve| 5 Applications of the proposed two-dimensionalization algorithm in other fields| 6 Summary and discussion| 7 Acknowledgement| 8 Reference

備考

内容記述タイプ

Other

内容記述

九州工業大学博士学位論文学位記番号:工博甲第589号学位授与年月日:令和6年3月25日

学位授与番号

甲第589号

学位名

博士（工学）

学位授与年月日

2024-03-25

学位授与機関

学位授与機関識別子Scheme

kakenhi

学位授与機関識別子

17104

学位授与機関名

九州工業大学

言語

学位授与年度

内容記述タイプ

Other

内容記述

令和5年度

出版タイプ

VoR

出版タイプResource

http://purl.org/coar/version/c_970fb48d4fbd8a85

アクセス権

open access

アクセス権URI

http://purl.org/coar/access_right/c_abf2

ID登録

10.18997/0002000700

ID登録タイプ

JaLC

戻る

views

See details

	Views

Versions

Ver.1

2024-05-28 00:28:03.769697

Show All versions

Cite as

Other

エクスポート

OAI-PMH

JPCOAR 2.0
JPCOAR 1.0
DublinCore
DDI

Other Formats

インデックスリンク

インデックスツリー

アイテム

Research on Two-Dimensionalization Algorithms for Improving Emotion Recognition Accuracy in Speech Data and its Evaluation of Generalized Deployment

× Zijun, Yang

Versions

Share

Cite as

Other

エクスポート

コミュニティ

メニューを最小化

インデックスリンク

インデックスツリー

アイテム

Research on Two-Dimensionalization Algorithms for Improving Emotion Recognition Accuracy in Speech Data and its Evaluation of Generalized Deployment

× Zijun, Yang

Versions

Share

Cite as

Other

エクスポート

コミュニティ