Install any skill in seconds. Free to start, no credit card required.
Get Started Free →智能语音转文字助手,基于 Qwen3-ASR 模型,支持实时语音识别和智能文本改写。可以将录音转换为文字,并一键改写成邮件、笔记、社交媒体文案,支持复制、分享和录音拼接。适用于会议纪要、语音备忘、内容创作等多种场景。
.claude/skills/anbeime-qwen3-asr-assistant/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-03 | ✗→✓ | ▲ Improved | 465% | 0% |
| case-07 | ✗→✓ | ▲ Improved | 122% | 0% |
| case-08 | ✗→✓ | ▲ Improved | 97% | 0% |
| case-10 | ✗→✓ | ▲ Improved | 83% | 0% |
| case-11 | ✗→✓ | ▲ Improved | 70% | 0% |
requests>=2.28.0 numpy>=1.21.0
python from scripts.asr_transcriber import Qwen3ASRTranscriber
transcriber = Qwen3ASRTranscriber() result = transcriber.transcribe( audio_file="recording.wav", language="zh-CN" ) text = result"text"]
python # 智能体处理拼接 full_text = original_text + "\n\n" + new_text
场景:录制会议语音,转换为结构化的会议笔记
执行流程:
python from scripts.asr_transcriber import Qwen3ASRTranscriber
transcriber = Qwen3ASRTranscriber() result = transcriber.transcribe( audio_file="meeting.wav", language="zh-CN" ) meeting_text = result"text"]
# 会议笔记
## 时间地点
## 参会人员
## 主要议题
场景:录制语音备忘,转换为正式邮件
执行流程:
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:
如有疑问,请及时联系我。
祝好! 张经理
场景:录制语音灵感,转换为社交媒体发布文案
执行流程:
☕️ 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
场景:录制长篇语音,分多段录音,最后拼接完整文本
执行流程:
python result1 = transcriber.transcribe("part1.wav") text1 = result1["text"]
python result2 = transcriber.transcribe("part2.wav") text2 = result2["text"]
python result3 = transcriber.transcribe("part3.wav") text3 = result3["text"]
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]
场景:语音转文字后,一键复制或分享到微信
执行流程:
文本已复制到剪贴板!
✅ 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
初始化:
pythonQwen3ASRTranscriber(api_key=None, base_url=None)
主要方法:
python# 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 }
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 5,063 | 7,905 | +56% | 1 | 1 | 0% | 836 | 3,621 | +333% | 0 | 0 | — |
case-02 | fail→fail | 3,100 | 10,532 | +240% | 1 | 1 | 0% | 507 | 3,731 | +636% | 0 | 0 | — |
case-03 | fail→pass | 4,515 | 11,693 | +159% | 1 | 1 | 0% | 759 | 4,292 | +465% | 0 | 0 | — |
case-04 | pass→pass | 14,738 | 13,802 | -6% | 1 | 1 | 0% | 3,246 | 4,647 | +43% | 0 | 0 | — |
case-05 | pass→pass | 18,096 | 18,095 | -0% | 1 | 1 | 0% | 3,523 | 5,934 | +68% | 0 | 0 | — |
case-06 | pass→pass | 19,667 | 17,750 | -10% | 1 | 1 | 0% | 3,735 | 5,884 | +58% | 0 | 0 | — |
case-07 | fail→pass | 5,841 | 2,509 | -57% | 1 | 1 | 0% | 1,214 | 2,698 | +122% | 0 | 0 | — |
case-08 | fail→pass | 7,022 | 1,706 | -76% | 1 | 1 | 0% | 1,256 | 2,476 | +97% | 0 | 0 | — |
case-09 | pass→pass | 7,765 | 1,942 | -75% | 1 | 1 | 0% | 1,569 | 2,590 | +65% | 0 | 0 | — |
case-10 | fail→pass | 8,572 | 2,641 | -69% | 1 | 1 | 0% | 1,460 | 2,668 | +83% | 0 | 0 | — |
case-11 | fail→pass | 7,956 | 2,076 | -74% | 1 | 1 | 0% | 1,518 | 2,585 | +70% | 0 | 0 | — |
case-12 | pass→pass | 12,361 | 12,020 | -3% | 1 | 1 | 0% | 2,204 | 3,905 | +77% | 0 | 0 | — |
case-13 | pass→pass | 9,231 | 7,199 | -22% | 1 | 1 | 0% | 1,735 | 3,477 | +100% | 0 | 0 | — |
case-14 | pass→pass | 11,765 | 13,741 | +17% | 1 | 1 | 0% | 2,036 | 4,388 | +116% | 0 | 0 | — |
case-15 | pass→pass | 13,589 | 9,581 | -29% | 1 | 1 | 0% | 1,934 | 3,919 | +103% | 0 | 0 | — |
case-16 | fail→pass | 11,618 | 10,232 | -12% | 1 | 1 | 0% | 1,990 | 4,011 | +102% | 0 | 0 | — |
case-17 | fail→pass | 9,644 | 1,959 | -80% | 1 | 1 | 0% | 1,777 | 2,480 | +40% | 0 | 0 | — |
case-18 | fail→pass | 12,701 | 1,921 | -85% | 1 | 1 | 0% | 131 | 2,530 | +1831% | 0 | 0 | — |
case-19 | fail→pass | 9,388 | 3,432 | -63% | 1 | 1 | 0% | 1,877 | 3,022 | +61% | 0 | 0 | — |
case-20 | pass→pass | 9,917 | 3,351 | -66% | 1 | 1 | 0% | 1,967 | 2,912 | +48% | 0 | 0 | — |
case-21 | pass→pass | 10,175 | 11,297 | +11% | 1 | 1 | 0% | 1,751 | 3,779 | +116% | 0 | 0 | — |
case-22 | pass→pass | 9,218 | 6,836 | -26% | 1 | 1 | 0% | 1,506 | 3,395 | +125% | 0 | 0 | — |
case-23 | pass→pass | 8,644 | 3,230 | -63% | 1 | 1 | 0% | 1,700 | 2,798 | +65% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 23 cases were attempted. The headline lift of +39 percentage points is the difference between those two pass rates over the 23 comparable cases.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.