本資料は、掲載誌(URI)等のリンク先にある学位授与機関のWebサイトやCiNii Dissertationsから、本文を自由に閲覧できる場合があります。
博士論文
国立国会図書館館内限定公開
収録元データベースで確認する
国立国会図書館デジタルコレクション
デジタルデータあり
公開元のウェブサイトで確認する
DOI[10.24561/00019784]のデータに遷移します
Statistical Signal Processing Approaches to Analysis and Synthesis of Bone-Conducted Speech
- 国立国会図書館永続的識別子
- info:ndljp/pid/12912228
国立国会図書館での利用に関する注記
資料に関する注記
一般注記:
- type:textIn recent years, much attention has been paid to bone-conducted (BC) speech for its superior noise-robustness and other potential features, t...
書店で探す
障害者向け資料で読む
全国の図書館の所蔵
国立国会図書館以外の全国の図書館の所蔵状況を表示します。
所蔵のある図書館から取寄せることが可能かなど、資料の利用方法は、ご自身が利用されるお近くの図書館へご相談ください
書店で探す
障害者向け資料で読む
書誌情報
この資料の詳細や典拠(同じ主題の資料を指すキーワード、著者名)等を確認できます。
デジタル
- 資料種別
- 博士論文
- 著者・編者
- 張, 詩銘
- 著者標目
- 出版年月日等
- 2022
- 出版年(W3CDTF)
- 2022
- 並列タイトル等
- 骨導音声の分析と合成への統計的信号処理アプローチ
- 授与機関名
- 埼玉大学
- 授与年月日
- 2022-03-24
- 授与年月日(W3CDTF)
- 2022-03-24
- 報告番号
- 甲第1239号
- 学位
- 博士(工学)
- 本文の言語コード
- eng
- 対象利用者
- 一般
- 一般注記
- type:textIn recent years, much attention has been paid to bone-conducted (BC) speech for its superior noise-robustness and other potential features, the use of BC speech is explored for speech processing. However, many properties of BC speech are still unclear in various regions in speech processing such as pitch extraction, speech synthesis, noise reduction, speech enhancement, and so on. In pitch extraction, some published papers indicate that BC speech can be used for its noise-robustness, however, these researches are not enough to clarify the properties and performance of BC speech for pitch extraction. In BC speech synthesis, there are very few works of BC speech synthesis as far as we know up to now. In the BC speech enhancement, there is still not a satisfactory approach to improve the quality of BC speech. Overall, many potential properties applications of BC speech are still unclear. In this paper, we propose several approaches to explore and analyze the performance, properties, and application of BC speech in pitch extraction and BC speech synthesis. Pitch (fundamental period) or fundamental frequency is one of the most important and needful parameters to represent a speech signal. The pitch extraction is conducted from a speech signal through an air-conducted (AC) microphone. AC speech is often used for pitch extraction in noisy environments, however, it is corrupted by noise. Therefore, it is difficult to extract the pitch information accurately from such noisy AC speech. We propose two noise-robust pitch extraction methods combining AC speech with BC speech in noise environments. For the first proposed method (WACF-CEP), the WACF method is implemented for the AC speech input, while the CEP method is done for the BC speech input, and then a multiplication of the processed AC speech and the processed BC speech is implemented in several types of noise environments. For the second proposed method (WACF-WACF), WACF is utilized to process the AC and BC speech signals, respectively, and then a multiplication of the processed AC speech and the processed BC speech is implemented in noisy environments. In both two proposed methods, multiplication from two functions and the use of BC speech accelerate the true peak enhancement, resulting in a decrease in pitch extraction errors. Experimental results show superior performances of both proposed methods relative to the state-of-the-art method in all the types of noise. Speech synthesis is an important region in speech processing. Many researchers explore the AC speech synthesis from the BC speech for the BC speech enhancement. However, there are very few works for BC speech synthesis. Experience indicates that BC speech is more suitable to be used in many regions for its clarified and potential properties than that of AC speech. For this reason, we propose a simple approach for BC speech synthesis from the corresponding AC speech in real-time. However, when we synthesize BC speech, we have no promising ways to do so. Although some speech synthesis techniques are well known, we cannot optimistically rely on them due to the lack of knowledge about the inherent properties of BC speech. Therefore, we investigate speech synthesis for BC speech. We propose a BC speech synthesis approach based on least squares method and conduct a pitch extraction experiment using the synthesized BC speech. From the identification results, a synthesis filter is designed in which AC speech is utilized as the input signal. BC speech is generated from the synthesis filter as the output signal. The synthesis method to be considered for BC speech is simple and straightforward and applicable to AC speech we usually use. Therefore, we can utilize the BC speech synthesis as a preprocessing part of AC speech processing, thereby an advanced implementation by BC speech could be exhibited. By experiments, it is shown that the BC speech synthesis method is satisfactory. Filter model, order selection, and stability in the methodology are discussed and noise-robustness gained by the BC speech synthesis is also investigated. A pitch extraction experiment is also implemented using the synthetic BC speech. The experimental results show that pitch extraction from the synthetic BC speech provides more accurate performance than that from the corresponding AC speech.Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131 Introduction 15 1.1 Production of Speech . . . . . . . . . . . . . . . . . . . . . . . . . 15 1.2 Speech Pyramid and Applications of Speech Processing . . . . . . . 16 1.3 Different Types of Speech and Microphone . . . . . . . . . . . . . . 18 1.4 Bone-Conducted Speech . . . . . . . . . . . . . . . . . . . . . . . . 19 1.5 Motivations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19 1.6 Organization of the Thesis . . . . . . . . . . . . . . . . . . . . . . . 202 Pitch Extraction Approaches for BC Speech 21 2.1 Background of Pitch Extraction . . . . . . . . . . . . . . . . . . . . 21 2.2 Conventional Pitch Extraction Methods . . . . . . . . . . . . . . . . 21 2.3 Methods in the Time Domain . . . . . . . . . . . . . . . . . . . . . 23 2.3.1 Auto-correlation Function (ACF) . . . . . . . . . . . . . . . 23 2.3.2 Average Magnitude Difference Function (AMDF) . . . . . . 24 2.3.3 Weighted Auto-correlation Function (WACF) . . . . . . . . 24 2.4 Methods in the Frequency Domain . . . . . . . . . . . . . . . . . . 26 2.4.1 Cepstrum (CEP) . . . . . . . . . . . . . . . . . . . . . . . . 26 2.5 Proposed Method (WACF-CEP) Combining AC and BC Speeches . 26 2.5.1 Properties of Proposed Method (WACF-CEP) . . . . . . . . 28 2.6 Proposed Method (WACF-WACF) Combining AC and BC Speeches 30 2.6.1 Properties of Proposed Method (WACF-WACF) . . . . . . . 33 2.7 Experiments for Pitch Extraction . . . . . . . . . . . . . . . . . . . 33 2.7.1 Noise Recording . . . . . . . . . . . . . . . . . . . . . . . . . 33 2.7.2 Speech Recording . . . . . . . . . . . . . . . . . . . . . . . . 34 2.7.3 SNR Calculation and Analysis . . . . . . . . . . . . . . . . . 35 2.7.4 Experiments of Proposed Method (WACF-CEP) . . . . . . . 36 2.7.5 Experiments of Proposed Method (WACF-WACF) . . . . . 41 2.7.6 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41 3 BC Speech Synthesis Based on Least Squares Method 55 3.1 Background of BC Speech Synthesis . . . . . . . . . . . . . . . . . . 55 3.2 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55 3.3 LS Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57 3.4 Speech Recording . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60 3.5 Filter Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61 3.6 Synthesizing BC Speech . . . . . . . . . . . . . . . . . . . . . . . . 65 3.7 Evaluation of the Synthesized BC Speech . . . . . . . . . . . . . . . 66 3.7.1 Listening Test . . . . . . . . . . . . . . . . . . . . . . . . . . 68 3.8 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70 3.9 Noise-Robustness Investigation . . . . . . . . . . . . . . . . . . . . 73 3.10 Pitch Extraction for Synthesized BC Speech . . . . . . . . . . . . . 75 4 Conclusion and Future Work 86指導教員 : 島村徹也教授
- DOI
- 10.24561/00019784
- 国立国会図書館永続的識別子
- info:ndljp/pid/12912228
- コレクション(共通)
- コレクション(障害者向け資料:レベル1)
- コレクション(個別)
- 国立国会図書館デジタルコレクション > デジタル化資料 > 博士論文
- 収集根拠
- 博士論文(自動収集)
- 受理日(W3CDTF)
- 2023-07-08T03:42:35+09:00
- 記録形式(IMT)
- application/pdf
- オンライン閲覧公開範囲
- 国立国会図書館内限定公開
- デジタル化資料送信
- 図書館・個人送信対象外
- 遠隔複写可否(NDL)
- 可
- 連携機関・データベース
- 国立国会図書館 : 国立国会図書館デジタルコレクション