Voice Transcriber Toolkit

Voice-to-Text Transcription Toolkit - 语音识别转文字，支持Whisper/Vosk引擎，批量处理，字幕导出 | Speech recognition & transcription with Whisper/Vosk engines, batch processing, su...

Security Scan

The skill's code, requirements, and runtime instructions are consistent with a local transcription toolkit; it requests no unrelated credentials or unusual system access, but it will install PyPI packages and may download large model weights at runtime.

Purpose & Capability

Name/description (Whisper/Vosk transcription, ffmpeg conversion, subtitle export) match the declared binaries (python3, ffmpeg), requirements.txt (openai-whisper, vosk), and the provided code. The functionality requested (audio conversion, transcription, subtitle export) is coherent with the assets included.

Instruction Scope

SKILL.md instructs installing Python deps and ffmpeg — appropriate for the stated purpose. The code does not read unrelated system files or contact unexpected external endpoints, but whisper.load_model(...) and the Whisper/Vosk libraries may download or load model weights from the network or local cache at runtime (large downloads). Example/test files insert an absolute sys.path ('/root/.openclaw/...') which is unusual but appears intended to help local execution in the platform workspace; it does not itself exfiltrate data but is worth noting.

Install Mechanism

Install is via pip (requirements.txt) and apt (ffmpeg). These are standard for this use case; pip packages will be fetched from PyPI which is normal but carries the usual dependency risk (supply-chain, version trust). No downloads from arbitrary URLs or archived/extracted remote payloads were found.

Credentials

The skill declares no required environment variables, no credentials, and no config paths. The code does not attempt to access secrets or unrelated environment variables. This is proportionate to a local transcription tool.

Persistence & Privilege

The skill does not request always:true and defaults allow user invocation/autonomous model invocation (platform default). It does not modify other skills or system-wide configs. No concerning persistence or privilege escalation behavior is present.

Assessment

This skill appears to do what it says: local transcription and audio conversion. Before installing, be aware that: (1) pip will install openai-whisper and vosk which may pull large model weights at runtime (network downloads, significant disk usage); (2) running whisper.load_model(...) will typically download models from the internet unless cached—plan for bandwidth and storage or pin small models explicitly; (3) the examples/tests add a hardcoded '/root/.openclaw/...' sys.path which is odd but not harmful—review/adjust paths for your environment; (4) as with any third-party pip package, review dependency versions and run in a sandbox if you have security concerns. No credentials or secret access are requested by this skill.

功能

• Transcribe — 单文件/批量音频转文字 (Whisper/Vosk)
• Convert — 音频格式转换 (ffmpeg)
• Export — 导出 SRT/VTT 字幕
• Info — 音频文件元信息提取

使用

from scripts.voice_transcriber import VoiceTranscriber, AudioConverter

transcriber = VoiceTranscriber(engine="whisper", model_size="base")

# 单文件转录
result = transcriber.transcribe("meeting.mp3", language="zh")
print(result["text"])

# 批量转录
results = transcriber.transcribe_batch(["file1.mp3", "file2.wav"])

# 导出字幕
srt = transcriber.export_subtitles(result, "srt")

# 音频转换
converter = AudioConverter()
converter.convert_to_wav("input.m4a", "output.wav")

CLI

python3 scripts/voice_transcriber.py