flykhan
2536c937e3
翻译自英文原版 maths-cs-ai-compendium，共 20 章全部完成。 第01章 向量 | 第02章 矩阵 | 第03章 微积分 第04章 统计学 | 第05章 概率论 | 第06章 机器学习 第07章 计算语言学 | 第08章 计算机视觉 | 第09章 音频与语音 第10章 多模态学习 | 第11章 自主系统 | 第12章 图神经网络 第13章 计算与操作系统 | 第14章 数据结构与算法 第15章 生产级软件工程 | 第16章 SIMD与GPU编程 第17章 AI推理 | 第18章 ML系统设计 第19章 应用人工智能 | 第20章 前沿人工智能 翻译说明： - 所有数学公式 $...$ / $$...$$、代码块、图片引用完整保留 - mkdocs.yml 配置中文导航 + language: zh - README.md 已翻译为中文（兼 docs/index.md） - docs/ 目录包含指向各章文件的 symlink - 约 29,000 行中文内容，排除 .cache/ 构建缓存
114 lines
8.4 KiB
XML
114 lines
8.4 KiB
XML
<svg xmlns="http://www.w3.org/2000/svg" width="700" height="290">
|
|
<defs>
|
|
<marker id="arr4" viewBox="0 0 10 6" refX="10" refY="3" markerWidth="8" markerHeight="6" orient="auto">
|
|
<polygon points="0,0 10,3 0,6" fill="#666"/>
|
|
</marker>
|
|
<marker id="arr4r" viewBox="0 0 10 6" refX="10" refY="3" markerWidth="8" markerHeight="6" orient="auto">
|
|
<polygon points="0,0 10,3 0,6" fill="#e74c3c"/>
|
|
</marker>
|
|
<marker id="arr4g" viewBox="0 0 10 6" refX="10" refY="3" markerWidth="8" markerHeight="6" orient="auto">
|
|
<polygon points="0,0 10,3 0,6" fill="#27ae60"/>
|
|
</marker>
|
|
</defs>
|
|
|
|
<!-- Title -->
|
|
<text x="350" y="22" text-anchor="middle" font-family="Arial, sans-serif" font-size="14" font-weight="bold" fill="#333">Voice Conversion with Speaker Disentanglement</text>
|
|
|
|
<!-- ===== TOP PATH: Source Speaker ===== -->
|
|
|
|
<!-- Source Audio -->
|
|
<rect x="20" y="50" width="95" height="50" rx="8" fill="#3498db" fill-opacity="0.15" stroke="#3498db" stroke-width="1.5"/>
|
|
<text x="67" y="70" text-anchor="middle" font-family="Arial, sans-serif" font-size="9" font-weight="bold" fill="#3498db">Source Audio</text>
|
|
<text x="67" y="82" text-anchor="middle" font-family="Arial, sans-serif" font-size="8" fill="#666">Speaker A</text>
|
|
<!-- Small waveform -->
|
|
<polyline points="30,90 38,86 46,94 54,84 62,93 70,85 78,92 86,87 94,91 102,88" fill="none" stroke="#3498db" stroke-width="0.7" stroke-opacity="0.4"/>
|
|
|
|
<!-- Arrow to encoder -->
|
|
<line x1="115" y1="75" x2="142" y2="75" stroke="#666" stroke-width="1.2" marker-end="url(#arr4)"/>
|
|
|
|
<!-- Content Encoder -->
|
|
<rect x="144" y="45" width="110" height="60" rx="8" fill="#9b59b6" fill-opacity="0.12" stroke="#9b59b6" stroke-width="1.5"/>
|
|
<text x="199" y="65" text-anchor="middle" font-family="Arial, sans-serif" font-size="9" font-weight="bold" fill="#9b59b6">Content Encoder</text>
|
|
<text x="199" y="78" text-anchor="middle" font-family="Arial, sans-serif" font-size="8" fill="#666">Extract linguistic</text>
|
|
<text x="199" y="89" text-anchor="middle" font-family="Arial, sans-serif" font-size="8" fill="#666">content features</text>
|
|
|
|
<!-- Arrow to content representation -->
|
|
<line x1="254" y1="75" x2="278" y2="75" stroke="#666" stroke-width="1.2" marker-end="url(#arr4)"/>
|
|
|
|
<!-- Content Representation (speaker-independent) -->
|
|
<rect x="280" y="45" width="120" height="60" rx="8" fill="#27ae60" fill-opacity="0.12" stroke="#27ae60" stroke-width="2"/>
|
|
<text x="340" y="63" text-anchor="middle" font-family="Arial, sans-serif" font-size="9" font-weight="bold" fill="#27ae60">Content</text>
|
|
<text x="340" y="76" text-anchor="middle" font-family="Arial, sans-serif" font-size="9" font-weight="bold" fill="#27ae60">Representation</text>
|
|
<text x="340" y="92" text-anchor="middle" font-family="Arial, sans-serif" font-size="8" fill="#666">(speaker-independent)</text>
|
|
|
|
<!-- Disentanglement visual: dashed line separating content from identity -->
|
|
<line x1="270" y1="120" x2="410" y2="120" stroke="#e74c3c" stroke-width="1.2" stroke-dasharray="6,3"/>
|
|
<text x="340" y="135" text-anchor="middle" font-family="Arial, sans-serif" font-size="8" font-weight="bold" fill="#e74c3c" font-style="italic">disentanglement boundary</text>
|
|
|
|
<!-- Content label above -->
|
|
<text x="310" y="115" text-anchor="middle" font-family="Arial, sans-serif" font-size="7" fill="#27ae60">WHAT is said</text>
|
|
<!-- Identity label below -->
|
|
<text x="370" y="148" text-anchor="middle" font-family="Arial, sans-serif" font-size="7" fill="#e74c3c">WHO says it</text>
|
|
|
|
<!-- ===== BOTTOM PATH: Target Speaker ===== -->
|
|
|
|
<!-- Target Speaker Audio -->
|
|
<rect x="20" y="158" width="95" height="50" rx="8" fill="#e74c3c" fill-opacity="0.15" stroke="#e74c3c" stroke-width="1.5"/>
|
|
<text x="67" y="176" text-anchor="middle" font-family="Arial, sans-serif" font-size="9" font-weight="bold" fill="#e74c3c">Target Audio</text>
|
|
<text x="67" y="188" text-anchor="middle" font-family="Arial, sans-serif" font-size="8" fill="#666">Speaker B</text>
|
|
<text x="67" y="200" text-anchor="middle" font-family="Arial, sans-serif" font-size="7" fill="#666">(reference)</text>
|
|
|
|
<!-- Arrow to speaker encoder -->
|
|
<line x1="115" y1="183" x2="142" y2="183" stroke="#666" stroke-width="1.2" marker-end="url(#arr4)"/>
|
|
|
|
<!-- Speaker Encoder -->
|
|
<rect x="144" y="153" width="110" height="60" rx="8" fill="#f39c12" fill-opacity="0.12" stroke="#f39c12" stroke-width="1.5"/>
|
|
<text x="199" y="173" text-anchor="middle" font-family="Arial, sans-serif" font-size="9" font-weight="bold" fill="#f39c12">Speaker Encoder</text>
|
|
<text x="199" y="186" text-anchor="middle" font-family="Arial, sans-serif" font-size="8" fill="#666">Extract speaker</text>
|
|
<text x="199" y="197" text-anchor="middle" font-family="Arial, sans-serif" font-size="8" fill="#666">identity embedding</text>
|
|
|
|
<!-- Arrow to speaker embedding -->
|
|
<line x1="254" y1="183" x2="278" y2="183" stroke="#666" stroke-width="1.2" marker-end="url(#arr4)"/>
|
|
|
|
<!-- Speaker Embedding -->
|
|
<rect x="280" y="158" width="120" height="50" rx="8" fill="#e74c3c" fill-opacity="0.12" stroke="#e74c3c" stroke-width="2"/>
|
|
<text x="340" y="178" text-anchor="middle" font-family="Arial, sans-serif" font-size="9" font-weight="bold" fill="#e74c3c">Speaker</text>
|
|
<text x="340" y="191" text-anchor="middle" font-family="Arial, sans-serif" font-size="9" font-weight="bold" fill="#e74c3c">Embedding</text>
|
|
<text x="340" y="203" text-anchor="middle" font-family="Arial, sans-serif" font-size="7" fill="#666">d-vector / x-vector</text>
|
|
|
|
<!-- ===== MERGE: Both paths into Decoder ===== -->
|
|
|
|
<!-- Arrow from content to decoder -->
|
|
<path d="M400,75 Q430,75 445,100 L460,115" fill="none" stroke="#27ae60" stroke-width="1.5" marker-end="url(#arr4g)"/>
|
|
<text x="418" y="86" font-family="Arial, sans-serif" font-size="7" fill="#27ae60">content</text>
|
|
|
|
<!-- Arrow from speaker embedding to decoder -->
|
|
<path d="M400,183 Q430,183 445,158 L460,143" fill="none" stroke="#e74c3c" stroke-width="1.5" marker-end="url(#arr4r)"/>
|
|
<text x="418" y="176" font-family="Arial, sans-serif" font-size="7" fill="#e74c3c">identity</text>
|
|
|
|
<!-- Decoder -->
|
|
<rect x="462" y="100" width="100" height="60" rx="8" fill="#9b59b6" fill-opacity="0.15" stroke="#9b59b6" stroke-width="1.5"/>
|
|
<text x="512" y="122" text-anchor="middle" font-family="Arial, sans-serif" font-size="10" font-weight="bold" fill="#9b59b6">Decoder</text>
|
|
<text x="512" y="136" text-anchor="middle" font-family="Arial, sans-serif" font-size="8" fill="#666">Synthesise mel /</text>
|
|
<text x="512" y="148" text-anchor="middle" font-family="Arial, sans-serif" font-size="8" fill="#666">waveform</text>
|
|
|
|
<!-- Arrow to output -->
|
|
<line x1="562" y1="130" x2="590" y2="130" stroke="#666" stroke-width="1.2" marker-end="url(#arr4)"/>
|
|
|
|
<!-- Converted Speech Output -->
|
|
<rect x="592" y="95" width="90" height="70" rx="8" fill="#27ae60" fill-opacity="0.18" stroke="#27ae60" stroke-width="1.5"/>
|
|
<text x="637" y="118" text-anchor="middle" font-family="Arial, sans-serif" font-size="9" font-weight="bold" fill="#27ae60">Converted</text>
|
|
<text x="637" y="131" text-anchor="middle" font-family="Arial, sans-serif" font-size="9" font-weight="bold" fill="#27ae60">Speech</text>
|
|
<text x="637" y="147" text-anchor="middle" font-family="Arial, sans-serif" font-size="8" fill="#666">Target voice</text>
|
|
<text x="637" y="158" text-anchor="middle" font-family="Arial, sans-serif" font-size="8" fill="#666">Source content</text>
|
|
|
|
<!-- Waveform squiggle on output -->
|
|
<polyline points="600,110 608,105 616,115 624,103 632,113 640,104 648,112 656,106 664,111 672,107" fill="none" stroke="#27ae60" stroke-width="0.7" stroke-opacity="0.4"/>
|
|
|
|
<!-- ===== NOTE BOX ===== -->
|
|
<rect x="20" y="230" width="660" height="45" rx="6" fill="#f5f5f5" stroke="#333" stroke-width="1"/>
|
|
<text x="35" y="247" font-family="Arial, sans-serif" font-size="9" font-weight="bold" fill="#333">Note:</text>
|
|
<text x="70" y="247" font-family="Arial, sans-serif" font-size="9" fill="#666">Key challenge: separate what is said (content) from who says it (identity). Approaches include information</text>
|
|
<text x="35" y="262" font-family="Arial, sans-serif" font-size="9" fill="#666">bottlenecks, instance normalisation, vector quantisation (VQ-VAE), and adversarial training to strip speaker</text>
|
|
<text x="35" y="275" font-family="Arial, sans-serif" font-size="9" fill="#666">information from the content encoder. Zero-shot VC uses a single reference utterance for unseen target speakers.</text>
|
|
</svg> |