Elevenlabs Transcribe

**Quick judgment**: This skill is a strong, production-leaning wrapper around ElevenLabs Scribe, offering high-quality multilingual transcription with flexible inputs (files, URLs, mic) and detailed outputs (JSON, diarization, events). It is best suited for environments where sending audio to ElevenLabs’ cloud API is acceptable and ffmpeg is available. **Main strengths** - High transcription quality with 90+ languages and optional speaker diarization. - Supports both batch and realtime streaming modes, including microphone input. - Rich JSON output (timestamps, language info, word-level data) suitable for downstream automation and analysis. - Agent-friendly features like `--quiet` mode and partial transcripts for streaming scenarios. **Key risks / limitations** - **External API dependency**: Requires a valid `ELEVENLABS_API_KEY` and stable network access; subject to rate limits, outages, and pricing changes. - **Privacy & compliance**: Audio is sent to ElevenLabs’ servers; not suitable for highly sensitive or regulated data unless policies explicitly allow this. - **Environment requirements**: Needs `ffmpeg` and a compatible shell/Python environment; may require extra setup or adaptation on Windows or constrained agent runtimes. - **Operational edge cases**: Non-zero exit on errors is good, but there’s no visible built-in retry/backoff or sophisticated error recovery. **Recommended scenarios** - Agents or backend services that need reliable, high-quality speech-to-text for calls, podcasts, or livestreams where cloud processing is acceptable. - Workflows needing structured transcription data (timestamps, speakers, events) for search, summarization, or analytics. - Interactive or streaming experiences (live captions, voicebots) that benefit from partial transcripts and quiet logging behavior. **Less suitable when** - You must keep all audio strictly on-prem or offline. - You cannot install `ffmpeg` or manage environment variables/API keys. - Extremely low-latency, on-device transcription is required without network dependence.