Singer Creation 创建歌手

A singer is a voice profile built from a reference audio clip and its preprocessed musical data. The SVS model uses the singer's voice characteristics to synthesize new singing. This page covers the complete singer creation workflow, including audio preprocessing.

歌手 是基于参考音频片段及其预处理后的音乐数据构建的声音配置文件。SVS 模型会利用歌手的语音特征来合成新的歌声。本页涵盖完整的歌手创建流程，包括音频预处理。

Overview 概览

The Singer Creator is a dedicated window for creating new singers. It has three sections: Basic Info (name and avatar), Reference Audio (the WAV file), and Audio Preprocessing (F0 and MIDI extraction). The right panel shows a live preview of how the singer will appear.

歌手创建器（Singer Creator）是用于创建新歌手的专用窗口。它包含三个部分：基本信息（名称和头像）、参考音频（WAV 文件）和 音频预处理（F0 与 MIDI 提取）。右侧面板会实时预览歌手的最终外观。

Reference Audio Requirements 参考音频要求

The reference audio is the most important part of singer creation. It tells the model what voice to imitate. Requirements:

参考音频是歌手创建中最关键的部分。它告诉模型要模仿的声音。要求如下：

Step-by-Step: Creating a Singer 分步操作：创建歌手

1. Open the Singer Creator 1. 打开歌手创建器

1. In the main window, click the + button in the singer panel header (left side).在主窗口中，点击歌手面板标题栏（左侧）的 + 按钮。
2. A dialog appears with two options: Open Singer Creator (create new) or Open Existing Singer File (load a .sxssinger file).弹出对话框，包含两个选项：Open Singer Creator（新建）或 Open Existing Singer File（加载 .sxssinger 文件）。
3. Click Open Singer Creator.点击 Open Singer Creator。

2. Enter Basic Info 2. 填写基本信息

• Singer Name (required): Used for display and file naming. Example: "MyVoice".Singer Name（必填）：用于显示和文件命名。例如："MyVoice"。
• Avatar: Choose Color (a solid color shown on the timeline track) or Image (upload a picture file).Avatar：选择 Color（时间线轨道上显示的纯色）或 Image（上传图片文件）。

3. Upload the WAV File 3. 上传 WAV 文件

1. Click the upload area (or drag a WAV file onto it).点击上传区域（或将 WAV 文件拖到该区域）。
2. Select a .wav file that meets the requirements above.选择符合上述要求的 .wav 文件。
3. After uploading, a waveform visualization appears. Click Preview to listen, or Clear to remove it.上传后会出现波形可视化。点击 Preview 试听，或点击 Clear 移除。

4. Trim if Longer Than 30 Seconds 4. 超过 30 秒则裁剪

If your WAV file exceeds 30 seconds, the WAV Trim dialog opens automatically:

如果 WAV 文件超过 30 秒，会自动弹出 WAV 裁剪对话框：

• Drag the selection overlay to choose which 30-second segment to use.拖动选区覆盖层，选择要使用的 30 秒片段。
• Drag the edges of the selection to adjust start/end.拖动选区边缘调整起始/结束位置。
• Enter exact values in the Start Position and Clip Length fields.在 Start Position 和 Clip Length 输入框中填入精确数值。
• Click Preview Clip to listen to the selected segment.点击 Preview Clip 试听所选片段。
• Click Confirm Trim to apply, or Cancel to discard the file entirely.点击 Confirm Trim 应用裁剪，或点击 Cancel 完全丢弃该文件。

5. Start Audio Preprocessing 5. 开始音频预处理

Preprocessing extracts the musical information (pitch, notes, timing) from your reference audio. This step is required — without it, the singer cannot be used for synthesis.

预处理会从参考音频中提取音乐信息（音高、音符、时序）。此步骤必填——没有它，歌手无法用于合成。

1. Click Start Audio Preprocessing. A new window opens showing the WAV waveform (top), MIDI editor (middle), and F0 curve (bottom).点击 Start Audio Preprocessing。新窗口打开，显示 WAV 波形（顶部）、MIDI 编辑器（中部）和 F0 曲线（底部）。

Follow the detailed preprocessing steps in the next section.

请按照下一节的详细预处理步骤操作。

6. Save the Singer File 6. 保存歌手文件

1. After preprocessing is complete, click Create & Save (✓) in the Singer Creator.预处理完成后，在歌手创建器中点击 Create & Save（✓）。
2. Choose a save location. The singer is saved as a .sxssinger file (JSON format with base64-encoded audio).选择保存位置。歌手以 .sxssinger 文件保存（JSON 格式，内含 base64 编码的音频）。
3. A success message confirms creation. The singer now appears in the main window's singer panel.成功消息确认创建完成。歌手随即出现在主窗口的歌手面板中。

Audio Preprocessing in Detail 音频预处理详解

The Audio Preprocessing window extracts MIDI notes and F0 (fundamental frequency) data from the reference audio. The SVS model uses this data to understand the singer's vocal characteristics.

音频预处理窗口会从参考音频中提取 MIDI 音符和 F0（基频）数据。SVS 模型利用这些数据来理解歌手的语音特征。

Step 1: Extract F0 (Pitch) 第 1 步：提取 F0（音高）

• Click RMVPE Extract F0. This uses the RMVPE neural model to detect the fundamental frequency curve from the audio.点击 RMVPE Extract F0。此操作使用 RMVPE 神经网络模型从音频中检测基频曲线。
• The F0 curve appears in the bottom panel. This is a read-only visualization of the detected pitch contour — you cannot edit it directly.F0 曲线显示在底部面板。这是检测到的音高轮廓的只读可视化——无法直接编辑。

Step 2: Extract MIDI Notes 第 2 步：提取 MIDI 音符

• Click Extract MIDI. This uses Basic Pitch (the recommended tool) to detect note boundaries and pitches.点击 Extract MIDI。此操作使用 Basic Pitch（推荐工具）检测音符边界和音高。
• Extracted notes appear on the MIDI canvas, and the note count is displayed.提取的音符显示在 MIDI 画布上，并显示音符数量。

Step 3: Edit MIDI Notes (Mandatory) 第 3 步：编辑 MIDI 音符（必填）

Check each note's pitch (does the MIDI pitch match the actual sung note?), note boundaries (do starts/ends align with the singing?), and note count (are all sung notes detected? are there false detections?). Edit operations:

检查每个音符的音高（MIDI 音高是否与实际演唱一致？）、音符边界（起始/结束是否与演唱对齐？）和音符数量（是否所有演唱音符都被检测到？是否有误检？）。编辑操作如下：

• Move notes: Drag up/down (pitch) or left/right (timing).移动音符：上下拖动（音高）或左右拖动（时序）。
• Resize notes: Drag the right edge.调整音符长度：拖动右边缘。
• Add notes: Click on empty space.添加音符：在空白处点击。
• Delete notes: Select and press Delete.删除音符：选中后按 Delete。
• Edit lyrics: Double-click a note. You must type the lyrics yourself — they are not auto-detected.编辑歌词：双击音符。歌词必须自己输入——系统不会自动识别歌词。
• One syllable over multiple pitches: Enter the lyric on the first note and enter the ASCII hyphen - on each following note. The editor converts those notes to continuation notes. Leave lyrics empty only for rests.一字多音高：在第一个音符填写歌词，后续音符输入半角连字符 -。编辑器会将后续音符转换为连音。只有休止音符才应留空歌词。

Step 4: Fill in Lyrics (Mandatory) 第 4 步：填写歌词（必填）

Every note must have a lyric. The preprocessing does not detect lyrics from audio. You must manually type them:

每个音符都必须有歌词。预处理不会从音频中识别歌词，必须手动输入：

• For Chinese singing: Enter Chinese characters (e.g., 我, 你, 好). You may append a digit 1–5 after a character to force a specific tone (e.g., 你2 好3), where 1–4 are the four tones and 5 is the neutral tone (轻声). Pinyin text is not accepted as Chinese — only Chinese characters are.中文演唱：输入汉字（如 我、你、好）。可在汉字后加数字 1–5 强制指定声调（如 你2 好3），其中 1–4 为四声，5 为轻声。不接受拼音作为中文输入——仅支持汉字。
• For English singing: Enter the English word being sung (e.g., hello, love).英文演唱：输入演唱的英文单词（如 hello、love）。
• For notes with no sung content (rests, breaths): Leave the lyric empty or use a space.没有演唱内容的音符（休止、换气）：歌词留空或输入空格。

Step 5: Save Preprocessed Data 第 5 步：保存预处理数据

1. Click Save (💾). This requires at least F0 or MIDI data to be extracted.点击 Save（💾）。此操作要求至少提取了 F0 或 MIDI 数据。
2. All preprocessed data is sent back to the Singer Creator window via IPC.所有预处理数据通过 IPC 发送回歌手创建器窗口。
3. The Singer Creator preview updates to show both "WAV ✓" and "Preprocess ✓" badges.歌手创建器预览会更新，显示 "WAV ✓" 和 "Preprocess ✓" 徽章。

Singer Preview 歌手预览

The right panel of the Singer Creator shows a live preview of how the singer will appear:

歌手创建器右侧面板会实时预览歌手的最终外观：

• Avatar (color or image)头像（颜色或图片）
• Singer name歌手名称
• WAV status badge (✓ when audio is loaded)WAV 状态徽章（加载音频后显示 ✓）
• Preprocess status badge (✓ when preprocessing is complete)预处理状态徽章（预处理完成后显示 ✓）

Tips for Good Reference Audio 优质参考音频的建议

• Use a clear, dry vocal recording with no reverb or echo.使用清晰、干涩的人声录音，不要混响或回声。
• Ensure no background noise (fans, air conditioning, room ambience).确保没有背景噪音（风扇、空调、房间环境音）。
• Include a varied pitch range — the model learns from what it hears.包含多变的音域范围——模型会从听到的内容中学习。
• Good articulation helps the model learn clearer phonemes.清晰的咬字有助于模型学习更清晰的音素。
• 10–30 seconds of continuous singing is ideal.10–30 秒的连续演唱最为理想。
• Use audio isolation tools (vocal removers) to extract clean vocals from mixed recordings if needed.如有必要，使用人声分离工具从混音中提取纯净人声。

Singer Management 歌手管理

Singer Status Indicators 歌手状态指示

• Normal: Singer loaded and ready.正常：歌手已加载，可用。
• ⚠ Singer file not found: The .sxssinger file was moved or deleted. Use Relocate to point to the new location.⚠ Singer file not found：.sxssinger 文件已被移动或删除。使用 Relocate 重新指向新位置。

Deleting a Singer 删除歌手

Right-click a singer and select Delete Singer. This removes the singer and all its fragments from the project.

右键点击歌手并选择 Delete Singer。这会从项目中移除该歌手及其所有片段。

Singer File Validation 歌手文件校验

When loading a .sxssinger file, SXSEditor validates file format/version compatibility, required fields (name, reference audio), and data integrity (MIDI notes, F0 data). Validation errors are shown in a report dialog; warnings indicate non-critical issues.

加载 .sxssinger 文件时，SXSEditor 会校验文件格式/版本兼容性、必填字段（名称、参考音频）和数据完整性（MIDI 音符、F0 数据）。校验错误会显示在报告对话框中；警告表示非关键问题。