subtitle-generation

字幕生成

使用 each::sense 为视频生成专业的字幕和说明文字。此技能可生成准确的转录文本、多语言字幕、动态字幕，以及适用于社交媒体、视频制作和无障碍场景的多种导出格式。

功能

• 自动生成字幕：自动将语音转录为文本，并提供准确的时间戳
• 多语言生成：根据音频生成多种语言的字幕
• 动态字幕：TikTok/Instagram 风格的逐词动态字幕
• SRT/VTT 导出：适用于编辑和分发的标准字幕格式
• 说话人分离：识别并标记对话中的不同说话人
• 字幕翻译：将现有字幕翻译成其他语言
• 硬字幕：将字幕直接渲染到视频中（硬编码）
• 卡拉 OK 风格：为音乐和歌词视频提供逐词高亮效果
• 时间调整：微调字幕时间和同步效果
• 批量处理：一次为多个视频生成字幕

快速开始

curl -X POST https://eachsense-agent.core.eachlabs.run/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "X-API-Key: $EACHLABS_API_KEY" \
  -H "Accept: text/event-stream" \
  -d '{
    "messages": [{"role": "user", "content": "Generate subtitles for this video with accurate timestamps"}],
    "model": "eachsense/beta",
    "stream": true,
    "mode": "max",
    "file_urls": ["https://example.com/my-video.mp4"]
  }'

字幕格式和输出

格式	扩展名	使用场景
SRT	.srt	通用格式，适用于大多数视频播放器和编辑器
VTT	.vtt	Web 视频、HTML5 播放器、YouTube
硬字幕	.mp4	社交媒体，无需播放器支持
JSON	.json	自定义应用程序、编程访问
ASS/SSA	.ass	高级样式、动漫字幕

使用场景示例

1. 从视频自动生成字幕

自动转录视频文件中的语音，并生成准确的逐词时间戳。

curl -X POST https://eachsense-agent.core.eachlabs.run/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "X-API-Key: $EACHLABS_API_KEY" \
  -H "Accept: text/event-stream" \
  -d '{
    "messages": [{"role": "user", "content": "Transcribe this video and generate subtitles with accurate timestamps. Output as SRT format. The video contains English speech."}],
    "model": "eachsense/beta",
    "stream": true,
    "mode": "max",
    "file_urls": ["https://example.com/interview-video.mp4"]
  }'

2. 生成多语言字幕

直接根据音频生成多种语言的字幕。

curl -X POST https://eachsense-agent.core.eachlabs.run/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "X-API-Key: $EACHLABS_API_KEY" \
  -H "Accept: text/event-stream" \
  -d '{
    "messages": [{"role": "user", "content": "Generate subtitles for this video in English, Spanish, and French. Provide separate SRT files for each language. The original audio is in English."}],
    "model": "eachsense/beta",
    "stream": true,
    "mode": "max",
    "file_urls": ["https://example.com/product-demo.mp4"]
  }'

3. 动态/样式化字幕（TikTok 风格）

创建醒目的动态字幕，并逐词高亮，这种样式在 TikTok 和 Instagram Reels 上很受欢迎。

curl -X POST https://eachsense-agent.core.eachlabs.run/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "X-API-Key: $EACHLABS_API_KEY" \
  -H "Accept: text/event-stream" \
  -d '{
    "messages": [{"role": "user", "content": "Add TikTok-style animated captions to this video. Use bold white text with black outline, word-by-word pop animation, centered at the bottom third of the screen. Make it trendy and engaging."}],
    "model": "eachsense/beta",
    "stream": true,
    "mode": "max",
    "file_urls": ["https://example.com/short-form-content.mp4"]
  }'

4. 导出 SRT/VTT

生成规范的标准格式字幕文件，供视频编辑器或流媒体平台使用。

curl -X POST https://eachsense-agent.core.eachlabs.run/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "X-API-Key: $EACHLABS_API_KEY" \
  -H "Accept: text/event-stream" \
  -d '{
    "messages": [{"role": "user", "content": "Transcribe this video and export subtitles in both SRT and VTT formats. Ensure proper line breaks (max 42 characters per line, 2 lines max). Include timestamps accurate to milliseconds."}],
    "model": "eachsense/beta",
    "stream": true,
    "mode": "max",
    "file_urls": ["https://example.com/documentary.mp4"]
  }'

5. 说话人分离（识别说话人）

生成能够识别并标记不同说话人的字幕，适用于对话、采访或播客。

curl -X POST https://eachsense-agent.core.eachlabs.run/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "X-API-Key: $EACHLABS_API_KEY" \
  -H "Accept: text/event-stream" \
  -d '{
    "messages": [{"role": "user", "content": "Generate subtitles for this podcast with speaker diarization. There are 2 speakers - identify them as Speaker 1 and Speaker 2 (or Host and Guest if you can determine roles). Format each line with the speaker label."}],
    "model": "eachsense/beta",
    "stream": true,
    "mode": "max",
    "file_urls": ["https://example.com/podcast-episode.mp4"]
  }'

6. 字幕翻译

将现有字幕从一种语言翻译成另一种语言，同时保留时间信息。

curl -X POST https://eachsense-agent.core.eachlabs.run/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "X-API-Key: $EACHLABS_API_KEY" \
  -H "Accept: text/event-stream" \
  -d '{
    "messages": [{"role": "user", "content": "Translate these English subtitles to Japanese. Preserve the original timing and format. Ensure natural Japanese phrasing rather than literal translation."}],
    "model": "eachsense/beta",
    "stream": true,
    "mode": "max",
    "file_urls": ["https://example.com/original-subtitles.srt"]
  }'

7. 硬字幕

将字幕直接渲染到视频文件中（硬编码/嵌入），无需播放器支持即可显示。

curl -X POST https://eachsense-agent.core.eachlabs.run/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "X-API-Key: $EACHLABS_API_KEY" \
  -H "Accept: text/event-stream" \
  -d '{
    "messages": [{"role": "user", "content": "Generate subtitles for this video and burn them directly into the video. Use white text with black background box, Arial font, positioned at bottom center. Output a new video file with embedded subtitles."}],
    "model": "eachsense/beta",
    "stream": true,
    "mode": "max",
    "file_urls": ["https://example.com/social-media-clip.mp4"]
  }'

8. 逐词卡拉 OK 风格字幕

创建带有逐词高亮效果的卡拉 OK 风格字幕，非常适合音乐视频和歌词内容。

curl -X POST https://eachsense-agent.core.eachlabs.run/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "X-API-Key: $EACHLABS_API_KEY" \
  -H "Accept: text/event-stream" \
  -d '{
    "messages": [{"role": "user", "content": "Create karaoke-style subtitles for this music video. Display lyrics with word-by-word highlighting as they are sung. Use a gradient color change effect (from white to yellow) for the currently sung word. Center the text on screen."}],
    "model": "eachsense/beta",
    "stream": true,
    "mode": "max",
    "file_urls": ["https://example.com/music-video.mp4"]
  }'

9. 调整字幕时间

微调字幕时间，使其与音频更好地同步。

# First, upload video and generate initial subtitles
curl -X POST https://eachsense-agent.core.eachlabs.run/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "X-API-Key: $EACHLABS_API_KEY" \
  -H "Accept: text/event-stream" \
  -d '{
    "messages": [{"role": "user", "content": "Generate subtitles for this video"}],
    "model": "eachsense/beta",
    "stream": true,
    "mode": "max",
    "session_id": "subtitle-timing-project",
    "file_urls": ["https://example.com/video-with-delay.mp4"]
  }'

# Then adjust timing in the same session
curl -X POST https://eachsense-agent.core.eachlabs.run/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "X-API-Key: $EACHLABS_API_KEY" \
  -H "Accept: text/event-stream" \
  -d '{
    "messages": [{"role": "user", "content": "The subtitles are appearing 500 milliseconds too early. Shift all subtitle timings forward by 500ms and regenerate the SRT file."}],
    "model": "eachsense/beta",
    "stream": true,
    "session_id": "subtitle-timing-project"
  }'

10. 批量生成字幕

在单个工作流中为多个视频生成字幕。

curl -X POST https://eachsense-agent.core.eachlabs.run/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "X-API-Key: $EACHLABS_API_KEY" \
  -H "Accept: text/event-stream" \
  -d '{
    "messages": [{"role": "user", "content": "Generate English subtitles for all these videos. Output SRT files for each. Use consistent formatting across all videos: max 2 lines, 42 characters per line, minimum 1 second display time per subtitle."}],
    "model": "eachsense/beta",
    "stream": true,
    "mode": "max",
    "file_urls": [
      "https://example.com/episode-01.mp4",
      "https://example.com/episode-02.mp4",
      "https://example.com/episode-03.mp4"
    ]
  }'

最佳实践

转录质量

• 清晰的音频：语音清晰且背景噪声较少时效果最佳
• 语言提示：指定源语言以提高准确性
• 说话人数：提供说话人数，以便更准确地进行说话人分离
• 上下文：提供内容背景（术语、姓名），以提高准确性

字幕格式

• 行长：每行不超过 42 个字符，便于阅读
• 显示时长：每条字幕应显示 1-7 秒
• 每条字幕的行数：每个字幕块最多 2 行
• 阅读速度：建议每分钟 150-180 个词，以便舒适阅读

动态字幕

• 字体选择：对于短视频内容，粗体无衬线字体效果最佳
• 对比度：使用描边或阴影，确保字幕在任何背景下都清晰可见
• 位置：避开平台 UI 元素，留出安全区域
• 动画：轻微动画比夸张效果更易阅读

翻译

• 文化适配：要求进行本地化，而不只是翻译
• 时间灵活性：有些语言需要更多阅读时间
• 字符限制：中日韩（CJK）语言每行通常可以使用更少的字符

字幕生成提示词建议

请求生成字幕时，请在提示词中包含以下详细信息：

1. 源语言：视频中说的是什么语言？
2. 目标格式：SRT、VTT、内嵌字幕还是动画字幕？
3. 样式要求：字体、颜色、位置、动画
4. 说话者信息：说话者人数；如已知，请注明角色
5. 特殊术语：专业词汇、名称、品牌
6. 输出语言：单一语言还是多种译文

提示词结构示例

"Generate [format] subtitles for this video.
Source language: [language].
[Number] speakers: [roles if known].
Style: [font, color, position requirements].
Special terms to recognize: [names, technical words].
Additional requirements: [line length, timing, etc.]"

模式选择

生成前先询问用户：

“您想要快速且经济，还是追求高质量？”

模式	适用场景	速度	质量
max	最终字幕、专业内容、对准确性要求极高的内容	较慢	最高
eco	快速草稿、审阅稿、批量处理	较快	良好

多轮字幕优化

使用 session_id 迭代优化字幕生成：

# Initial subtitle generation
curl -X POST https://eachsense-agent.core.eachlabs.run/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "X-API-Key: $EACHLABS_API_KEY" \
  -H "Accept: text/event-stream" \
  -d '{
    "messages": [{"role": "user", "content": "Generate subtitles for this video with speaker identification"}],
    "model": "eachsense/beta",
    "stream": true,
    "session_id": "subtitle-project-001",
    "file_urls": ["https://example.com/interview.mp4"]
  }'

# Refine based on feedback
curl -X POST https://eachsense-agent.core.eachlabs.run/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "X-API-Key: $EACHLABS_API_KEY" \
  -H "Accept: text/event-stream" \
  -d '{
    "messages": [{"role": "user", "content": "Change Speaker 1 label to John and Speaker 2 to Sarah. Also fix the spelling of TensorFlow wherever it appears."}],
    "model": "eachsense/beta",
    "stream": true,
    "session_id": "subtitle-project-001"
  }'

# Add styling and export
curl -X POST https://eachsense-agent.core.eachlabs.run/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "X-API-Key: $EACHLABS_API_KEY" \
  -H "Accept: text/event-stream" \
  -d '{
    "messages": [{"role": "user", "content": "Now create a burned-in version with the corrected subtitles. Use yellow text for John and cyan for Sarah."}],
    "model": "eachsense/beta",
    "stream": true,
    "session_id": "subtitle-project-001"
  }'

语言支持

each::sense 支持生成 50 多种语言的字幕，包括：

语言	代码	备注
英语	en	美国、英国、澳大利亚变体
西班牙语	es	拉丁美洲和欧洲变体
法语	fr	法国和加拿大变体
德语	de
日语	ja
韩语	ko
中文	zh	简体和繁体
阿拉伯语	ar	支持 RTL
印地语	hi
葡萄牙语	pt	巴西和欧洲变体

错误处理

错误	原因	解决方案
Failed to create prediction: HTTP 422	余额不足	在 eachlabs.ai 充值
转录质量低	音频质量差	提供更清晰的音频源
语言检测失败	混合语言或语音不清晰	明确指定源语言
超时	视频较长或处理复杂	将客户端超时设置为至少 10 分钟

客户端配置

重要：较长视频的字幕生成可能需要较长时间。

• 最短超时时间：10 分钟（600 秒）
• 建议：根据视频长度设置超时时间（每分钟视频时长设置 2–3 分钟）
• 流式传输：使用 SSE 事件处理来显示进度

相关技能

• each-sense - 核心 API 文档
• video-generation - 生成带内置字幕的视频
• voice-audio - 音频处理和语音合成
• video-edit - 视频编辑和后期制作