会议转录

会议记录模板

meeting_transcript:
  metadata:
    title: "{{meeting_title}}"
    date: "{{date}}"
    duration: "{{duration}}"
    attendees: "{{speakers}}"
    
  output_template: |
    # {{title}}
    
    **日期:** {{date}}
    **时长:** {{duration}}
    **参会者:** {{attendees}}
    
    ## 摘要
    {{ai_summary}}
    
    ## 关键点
    {{#each key_points}}
    - {{this}}
    {{/each}}
    
    ## 行动项
    {{#each action_items}}
    - [ ] {{task}} - @{{assignee}} - 截止日期: {{due_date}}
    {{/each}}
    
    ## 完整转录
    {{#each segments}}
    **[{{timestamp}}] {{speaker}}:** {{text}}
    
    {{/each}}

说话人分离

diarization_config:
  min_speakers: 2
  max_speakers: 10
  
  speaker_labels:
    - name: "Speaker 1"
      voice_sample: "sample_1.wav"  # 可选
    - name: "Speaker 2"
      voice_sample: "sample_2.wav"
      
  output_format:
    speaker_prefix: true
    speaker_timestamps: true
    
  example_output: |
    [00:00:05] SPEAKER_1: 欢迎大家参加今天的会议。
    [00:00:12] SPEAKER_2: 感谢邀请我们。
    [00:00:18] SPEAKER_1: 让我们从议程开始。

字幕生成

SRT格式

subtitle_config:
  format: srt
  
  timing:
    max_duration: 7  # 每条字幕的秒数
    min_gap: 0.1     # 字幕间隔秒数
    chars_per_line: 42
    max_lines: 2
    
  style:
    case: sentence  # sentence, upper, lower
    numbers: words  # words, digits
    
  example_output: |
    1
    00:00:05,000 --> 00:00:08,500
    欢迎参加今天的演示
    关于转录自动化。
    
    2
    00:00:09,000 --> 00:00:12,000
    让我从解释基本概念开始。

VTT格式

vtt_config:
  format: vtt
  
  features:
    cue_settings: true
    styling: true
    
  example_output: |
    WEBVTT
    
    00:00:05.000 --> 00:00:08.500 align:center
    欢迎参加今天的演示
    关于转录自动化。
    
    00:00:09.000 --> 00:00:12.000 align:center
    <v Speaker 1>让我从解释基本概念开始。

集成工作流

Zoom集成

zoom_transcription:
  trigger:
    event: recording_completed
    
  workflow:
    - step: download_recording
      source: zoom_cloud
      
    - step: transcribe
      engine: whisper
      language: auto
      
    - step: diarize
      identify_speakers: true
      
    - step: generate_notes
      template: meeting_notes
      include_summary: true
      extract_action_items: true
      
    - step: distribute
      destinations:
        - notion_page
        - slack_channel
        - email_attendees

YouTube集成

youtube_subtitles:
  trigger:
    event: video_uploaded
    
  workflow:
    - step: download_audio
      source: youtube_video
      
    - step: transcribe
      engine: whisper
      task: transcribe
      
    - step: generate_subtitles
      formats: [srt, vtt]
      
    - step: translate
      target_languages: [es, zh, ja, de, fr]
      
    - step: upload_subtitles
      destination: youtube
      as_cc: true

播客处理

podcast_workflow:
  input:
    source: rss_feed
    format: audio/mp3
    
  processing:
    - transcribe:
        engine: whisper
        model: large
        
    - generate_chapters:
        detect_topics: true
        min_duration: 60  # 秒
        
    - create_show_notes:
        summarize: true
        extract_links: true
        highlight_quotes: true
        
    - create_searchable_index:
        full_text: true
        timestamps: true
        
  output:
    - transcript_txt
    - chapters_json
    - show_notes_md
    - search_index

语言支持

多语言转录

multilingual:
  auto_detect: true
  
  supported_languages:
    - code: en
      name: English
      model: large
      
    - code: zh
      name: Chinese
      model: large
      
    - code: es
      name: Spanish
      model: large
      
    - code: ja
      name: Japanese
      model: medium
      
  translation:
    enabled: true
    target: en
    preserve_original: true

语码转换

code_switching:
  enabled: true
  primary_language: en
  secondary_languages: [zh, es]
  
  output: |
    [00:01:23] The next topic is about 人工智能,
    which has been muy importante in recent years.
    
  handling:
    detect_language_per_segment: true
    tag_language_switches: true

质量增强

后处理

post_processing:
  text_cleanup:
    - remove_filler_words: ["um", "uh", "like"]
    - fix_common_errors: true
    - normalize_numbers: true
    
  formatting:
    - add_punctuation: true
    - capitalize_sentences: true
    - paragraph_breaks: true
    
  speaker_attribution:
    - merge_short_segments: true
    - min_segment_duration: 1.0
    
  output_enhancement:
    - add_timestamps: true
    - highlight_keywords: true
    - generate_summary: true

准确性指标

转录质量报告
═══════════════════════════════════════

文件: meeting_2024_01_15.mp3
时长: 45:32
引擎: Whisper Large

指标:
词错误率 (WER):  4.2%
字符错误率:   2.8%
置信度分数:       0.94

说话人分离:
检测到的说话人: 4
分离准确率: 91%

处理时间:
总计: 8分23秒
实时因子: 0.18倍

检测到的问题:
• 12:34处置信度低（背景噪音）
• 23:45处语音重叠
• 34:12处未知说话人

API示例

OpenAI Whisper

import openai

# 转录音频
with open("meeting.mp3", "rb") as audio_file:
    transcript = openai.Audio.transcribe(
        model="whisper-1",
        file=audio_file,
        response_format="verbose_json",
        timestamp_granularities=["word", "segment"]
    )

# 访问结果
for segment in transcript.segments:
    print(f"[{segment.start:.2f}] {segment.text}")

AssemblyAI

import assemblyai as aai

transcriber = aai.Transcriber()

config = aai.TranscriptionConfig(
    speaker_labels=True,
    auto_chapters=True,
    entity_detection=True
)

transcript = transcriber.transcribe(
    "https://example.com/meeting.mp3",
    config=config
)

for utterance in transcript.utterances:
    print(f"Speaker {utterance.speaker}: {utterance.text}")

最佳实践

1. 高质量音频 : 干净的输入 = 更好的输出
2. 选择合适模型 : 平衡速度与准确性
3. 使用说话人分离 : 清晰识别说话人
4. 后处理 : 清理自动化输出
5. 验证关键内容 : 重要内容人工审核
6. 考虑隐私 : 处理敏感内容
7. 高效存储 : 压缩和索引
8. 提供上下文 : 词汇提示有帮助

每周安装次数

0

仓库

claude-office-s…s/skills

GitHub星标数

7

首次出现

1970年1月1日

安全审计

Gen Agent Trust HubPassSocketPassSnykWarn