Pronunciation Coach

Pronunciation coaching with real voice analysis using Azure Speech Services. Analyzes audio files for phoneme-level accuracy, fluency, prosody, and intonatio...

This skill appears to do what it claims: it will read audio files from ~/.openclaw/media/inbound/, convert them (ffmpeg), and upload them to Microsoft Azure Speech for pronunciation assessment. Before installing: 1) Confirm you are comfortable sending users' audio to Microsoft (privacy and billing/usage matters). 2) Provide an Azure Speech key and region via AZURE_SPEECH_KEY and AZURE_SPEECH_REGION. 3) Ensure ffmpeg and Node.js are available in the agent environment. 4) Note the SKILL.md suggests sending results back to users (text and TTS) but the skill does not implement Telegram messaging or TTS — you will need the agent or other skills to have those permissions. 5) Fix the registry metadata mismatch (it should declare required env vars) and verify the skill's source/homepage if provenance matters. If you need stronger assurance, review the scripts locally or run them in a sandboxed environment before granting access to real user audio or credentials.

Type: OpenClaw Skill Name: pronunciation-coach Version: 1.0.4 The skill is benign, transparently declaring its purpose to analyze audio using Azure Speech Services. It explicitly requests read access to `~/.openclaw/media/inbound/` and network access to `*.stt.speech.microsoft.com` in `skill.json`. The `pronunciation-assess.sh` script demonstrates good security practices by implementing checks against filename option injection (`case "$AUDIO_FILE" in -*)`, `ffmpeg -i -- "$AUDIO_FILE"`) and sanitizing the `REFERENCE_TEXT` to prevent JSON injection before sending it to Azure. The `SKILL.md` instructions guide the agent to perform its stated function without any evidence of prompt injection attempts to subvert its behavior or access unauthorized data.