Audio Recognition

音频语音识别服务(Speech-to-Text)。当用户上传音频文件，需要将语音内容转换为文字，或需要识别音频中的特定信息（如关键词、歌曲名）时触发。 适用于：(1) 会议录音转写 (2) 音频内容提取 (3) 语音指令识别 (4) 音视频字幕生成

Security Scan

The skill is an instruction-only description of an audio speech-to-text pipeline and its requirements are consistent with that purpose; it does not request credentials, install software, or instruct actions outside its stated scope.

Purpose & Capability

The name/description (speech-to-text, diarization, punctuation, multi-language) aligns with the SKILL.md content. The skill does not ask for unrelated credentials, binaries, or config paths.

Instruction Scope

The SKILL.md outlines preprocessing, feature extraction, ASR models (Whisper/WeNet/Paraformer), and postprocessing at a high level. It does not tell the agent to read unrelated files or exfiltrate data, but it is high-level and leaves implementation choices unspecified (e.g., which model/service to call), so runtime behavior depends on the agent environment and any integrations the agent has.

Install Mechanism

No install specification or code files are present — the skill is instruction-only, so nothing will be written or executed by default.

Credentials

The skill requires no environment variables or credentials as declared, which is coherent for a descriptive spec. However, real implementations often require API keys or local model binaries; the SKILL.md does not request them or describe secure handling, so users should verify how the agent will implement model calls.

Persistence & Privilege

always:false and default model invocation settings are used. The skill does not request persistent presence or system-level configuration changes.

Scan Findings in Context

[no-findings] expected: Regex scanner found no code files to analyze. This is expected because the skill is instruction-only (SKILL.md).

Assessment

This skill is a high-level spec for an audio speech-to-text pipeline and appears coherent, but it does not implement anything by itself. Before installing or enabling: (1) confirm which runtime or service the agent will actually use (local model vs cloud provider); (2) if it uses third-party cloud APIs, expect to need API keys and verify how audio is uploaded and stored — the SKILL.md's privacy promise is descriptive but not enforceable; (3) verify accuracy claims (95% for Mandarin) against your expected audio conditions; and (4) ensure you have legal/consent coverage for processing any sensitive audio.

音频语音识别 (Audio Recognition)

将音频中的语音准确转写为文字，并能区分不同说话人。

核心能力

• 语音转文字 (Speech-to-Text)
• 说话人分离 (Diarization)
• 标点与断句优化
• 支持多语言识别

工作流程

1. 音频预处理

• 降噪处理
• 格式转换 (统一为16kHz/16bit PCM或MP3)
• 音量标准化

2. 声学特征提取

• 提取MFCC、FBANK等声学特征
• 用于后续ASR模型输入

3. ASR语音识别

• 使用ASR模型进行语音识别
• 生成文字初稿
• 支持模型：Whisper、WeNet、Paraformer等

4. 后处理

• 文字纠错
• 断句与标点添加
• 说话人分离标注

5. 输出结果

• 最终识别文本
• 说话人标签（如需要）
• 时间戳（如需要）

质量目标

• 准确率：95%以上（标准普通话录音）
• 实时性：支持实时和离线两种模式

限制说明

1. 噪音限制：背景噪音过大时识别效果下降
2. 口音限制：重度方言/口音可能影响准确率
3. 隐私保护：用户上传的音频仅用于本次识别，不得用于模型训练或其他用途
4. 语义理解：仅负责语音转文字，不理解文本语义内容

参考服务

• 讯飞听见
• Google Speech-to-Text
• 阿里云语音识别
• 腾讯云语音识别

适用场景

• 会议纪要自动生成
• 音频/视频字幕制作
• 语音内容检索
• 录音文件整理