| DC 欄位 |
值 |
語言 |
| DC.contributor | 電機工程學系 | zh_TW |
| DC.creator | 周軒宇 | zh_TW |
| DC.creator | Xuan-Yu Chou | en_US |
| dc.date.accessioned | 2023-4-21T07:39:07Z | |
| dc.date.available | 2023-4-21T07:39:07Z | |
| dc.date.issued | 2023 | |
| dc.identifier.uri | http://ir.lib.ncu.edu.tw:444/thesis/view_etd.asp?URN=109521031 | |
| dc.contributor.department | 電機工程學系 | zh_TW |
| DC.description | 國立中央大學 | zh_TW |
| DC.description | National Central University | en_US |
| dc.description.abstract | 移動多媒體電子設備的尺寸和厚度減小，嚴格限制了可配置揚聲器的尺
寸。因此，很容易犧牲播放的音質，尤其是低音。在本文中，我們提出了一
種基於神經網路的虛擬低音增強系統來解決這項問題。此外，虛擬低音增強
任務中產生的額外諧波可能會導致算術溢出而發生削波失真。因此，我們在
系統末端添加了一個多頻段壓縮器，以減少由於虛擬低音增強而導致的削
波失真。虛擬低音增強可以分為兩種主要的方法，一種是非線性元件（Nonlinear Device, NLD），另一種則是相位聲碼器（Phase Vocoder, PV）。NLD 通
過非線性元件（如乘法迴圈）直接在時域中產生諧波來實現虛擬低音; 而 PV
首先將訊號轉換至頻域，並使用頻譜偏移產生更高次的諧波。相較之下，由
於其設計特性，NLD 更適合使用於鼓和打擊樂等瞬態訊號(transient signal)，
而 PV 更適合人聲等穩態訊號(stationary signal)。因此，我們首先使用神經
網路將輸入音訊訊號分離成瞬態和穩態分量，並分別對它們應用虛擬低音
增強的方法，我們使用這些技術提出了一個完整的虛擬低音增強系統。最後，
通過與其他虛擬低音系統相比的主觀聽覺測試，可以驗證我們的虛擬低音
系統具有更高的低音感知和更低的失真。
| zh_TW |
| dc.description.abstract | The reduced size and thickness of mobile multimedia electronics strictly limit
the size of the configurable loudspeakers. As a result, it is easy to sacrifice the
sound quality of playback, especially the bass. In this paper, we propose a neural
network-based virtual bass system to solve this problem. In addition, the
additional harmonics generated in the virtual bass enhancement may lead to
arithmetic overflow and distortion due to clipping. Therefore, we add a multiband
compressor at the end of the system to reduce clipping due to virtual bass
enhancement. Virtual bass enhancement can be divided into two main approaches,
one is the non-linear device (NLD), and the other is the phase vocoder (PV). NLD
achieves virtual bass by generating harmonics directly in the time domain through
a non-linear device such as a multiplication loop. The PV first converts the signal
to the frequency domain and uses spectrum shifting to generate higher harmonics.
In contrast, due to their design characteristics, NLDs are more suitable for
transient signals such as drums and percussion, while PVs are more suitable for
stationary signals such as vocals. Therefore, we first use a neural network to split
the input audio signal into transient and stationary components and apply the
virtual bass enhancement to them separately. We use these techniques to propose
a complete virtual bass enhancement system. Finally, using subjective listening
tests compared with other virtual bass systems, we can see that our virtual bass
system has higher bass perception and lower distortion. | en_US |
| DC.subject | 虛擬低音系統 | zh_TW |
| DC.subject | 音訊瞬態/穩態分離 | zh_TW |
| DC.subject | 深度神經網路 | zh_TW |
| DC.subject | 非線性元件 | zh_TW |
| DC.subject | 相位聲碼器 | zh_TW |
| DC.subject | 多頻段壓縮器 | zh_TW |
| DC.subject | Virtual bass system | en_US |
| DC.subject | Stationary-transient source separation | en_US |
| DC.subject | Deep neural networks | en_US |
| DC.subject | Nonlinear device | en_US |
| DC.subject | Phase vocoder | en_US |
| DC.subject | Multiband compressor | en_US |
| DC.title | 具有基於神經網路的音訊瞬態/穩態分離和多頻 段壓縮器之虛擬低音系統 | zh_TW |
| dc.language.iso | zh-TW | zh-TW |
| DC.title | Virtual bass system with neural network based transient/stationary audio separation and multiband compressor | en_US |
| DC.type | 博碩士論文 | zh_TW |
| DC.type | thesis | en_US |
| DC.publisher | National Central University | en_US |