qwen3-asr-assistant

Qwen3-ASR Intelligent Speech-to-Text Assistant

Task Objectives

This Skill is used for: converting speech to text and providing intelligent text rewriting functionality
Capabilities include:
- Real-time speech recognition (speech-to-text)
- Intelligent text rewriting (emails, notes, social media copy)
- Text concatenation (merging multiple recordings)
- One-click copy and share
Trigger conditions: when the user requests "speech-to-text", "recording-to-text", "voice memo", etc.

Prerequisites

Dependency description: Python libraries required for Qwen3-ASR invocation
```
requests>=2.28.0
numpy>=1.21.0
```
No additional files or folder preparation required

Operation Steps

Standard process (speech-to-text + intelligent rewriting)

Record/Upload audio(provided by the caller)
- Click the record button to start recording
- Click stop to end recording
- Or upload an existing audio file

Speech-to-text(the agent invokes the script)

from scripts.asr_transcriber import Qwen3ASRTranscriber

transcriber = Qwen3ASRTranscriber()
result = transcriber.transcribe(
    audio_file="recording.wav",
    language="zh-CN"
)
text = result["text"]

Text rewriting(processed by the agent)
- Select the rewriting type according to user needs:
  - Rewrite as email: formal and structured, including subject, body, and signature
  - Rewrite as notes: clear key points and well-organized, using lists and markers
  - Rewrite as social media copy: concise and engaging, using emojis and hashtags
- The agent analyzes the source content and identifies key information
- Adjusts tone, structure, and style according to the rewriting type
Copy/Share(processed by the agent)
- One-click copy: the agent copies the rewritten text to the clipboard
- One-click share: the agent generates a sharing format suitable for platforms such as WeChat
Recording splicing(processed by the agent)
- Continue recording to generate new text
- The agent splices the new text onto the original text
- Maintains text coherence and adds appropriate connectives

Multi-recording splicing process

First recording: perform speech-to-text according to the standard process
Continue recording: the user clicks continue recording
Convert to text: call the script to recognize the new recording

Intelligent splicing: the agent splices the new text onto the original text

# 智能体处理拼接
full_text = original_text + "\n\n" + new_text

Resource Index

Required scripts:
- scripts/asr_transcriber.py(Purpose: speech-to-text, supports multiple audio formats and real-time recognition)
Domain references:
- references/asr-api-config.md(When to read: when you need to understand ASR API configuration and parameters)
- references/text-rewrite-guide.md(When to read: when you need to understand text rewriting rules and templates)
Output asset: No preset template; the rewritten text is dynamically generated according to requirements

Notes

Recording quality: Ensure the recording is clear, with no excessive background noise, to improve recognition accuracy
Text rewriting: Make full use of the agent's language understanding and creation capabilities, and avoid writing scripts for rewriting
Concatenation coherence: When concatenating multiple recording segments, the agent should add appropriate connectives to keep the text coherent
Rewriting accuracy: When rewriting, the core information and intent of the original text should be preserved, and content should not be arbitrarily added or removed
Scenario adaptation: Choose the appropriate rewriting type according to the usage scenario (email/notes/social media)

Usage examples

Example 1: Meeting minutes to notes

Scenario: Record meeting audio and convert it into structured meeting notes

Execution process：

Record meeting audio (60 minutes)

Call the script to convert to text:

from scripts.asr_transcriber import Qwen3ASRTranscriber

transcriber = Qwen3ASRTranscriber()
result = transcriber.transcribe(
    audio_file="meeting.wav",
    language="zh-CN"
)
meeting_text = result["text"]

The agent rewrites it into notes:

# 会议笔记

## 时间地点
- 时间：2024年1月23日
- 地点：公司会议室

## 参会人员
- 张经理、李主管、王工程师

## 主要议题
1. Q1业绩回顾
   - 销售额增长20%
   - 新用户增加5000人

2. Q2工作计划
   - 推出新产品线
   - 加强市场推广
   - 优化客户服务

3. 决议事项
   - 批准新产品线预算
   - 成立市场推广小组

Example 2: Voice to email

Scenario: Record a voice memo and convert it into a formal email

Execution process：

Record a voice memo: "Xiao Wang, regarding tomorrow's project meeting, please prepare the following materials: project progress report, budget sheet, and team member list. The meeting time is 10 a.m., and the location is Meeting Room 3."
Call the script to convert to text

The agent rewrites it into an email:

主题：明天项目会议准备材料

小王：

你好！

关于明天上午10点在3号会议室召开的项目会议，请提前准备好以下材料：

1. 项目进度报告
2. 预算表
3. 团队人员名单

如有疑问，请及时联系我。

祝好！
张经理

Example 3: Voice to social media copy

Scenario: Record voice inspiration and convert it into social media post copy

Execution process：

Record voice inspiration: "Today I tried an amazing coffee shop. The coffee was very rich, and the environment was very comfortable, very suitable for working. I recommend it to everyone!"
Call the script to convert to text

The agent rewrites it into social media copy:

☕️ 今日份咖啡推荐！

今天发现了一家宝藏咖啡店 ☕✨

咖啡口感浓郁，环境舒适超治愈，简直是工作充电的好地方～

#咖啡探店 #工作日常 #周末好去处

📍 地址：[咖啡店名称]
⭐ 推荐：招牌拿铁、手冲咖啡

Example 4: Concatenating multiple recording segments

Scenario: Record a long voice message in multiple segments, and finally concatenate the complete text

Execution process：

First recording segment(0-10 minutes):

result1 = transcriber.transcribe("part1.wav")
text1 = result1["text"]

Second recording segment(10-20 minutes):

result2 = transcriber.transcribe("part2.wav")
text2 = result2["text"]

Third recording segment(20-30 minutes):

result3 = transcriber.transcribe("part3.wav")
text3 = result3["text"]

Agent concatenation：

完整文本：

[第一段内容]

...（智能体添加连接词）...

[第二段内容]

...（智能体添加连接词）...

[第三段内容]

Example 5: One-click copy and share

Scenario: After voice-to-text, copy or share to WeChat with one click

Execution process：

Voice-to-text
The agent rewrites it into the target format
One-click copy：
```
文本已复制到剪贴板！
```

One-click share to WeChat：

✅ 文本已生成，可以分享到微信

分享格式：
[改写后的文本]

#会议纪要 #工作效率

API reference

Qwen3ASRTranscriber class

Initialization：

Qwen3ASRTranscriber(api_key=None, base_url=None)

Main methods：

# 语音转文字
transcribe(
    audio_file: str,
    language: str = "zh-CN",
    format: str = "wav",
    sample_rate: int = 16000,
    return_timestamps: bool = False
) -> dict

# 返回格式
{
    "success": True,
    "text": "识别的文字",
    "language": "zh-CN",
    "duration": 120.5,
    "segments": [...]  # 如果 return_timestamps=True
}

About this skill

Task Objectives

Prerequisites

Operation Steps

Standard process (speech-to-text + intelligent rewriting)

Multi-recording splicing process

Resource Index

Notes

Usage examples

Example 1: Meeting minutes to notes

Example 2: Voice to email

Example 3: Voice to social media copy

Example 4: Concatenating multiple recording segments

Example 5: One-click copy and share

API reference

Qwen3ASRTranscriber class