Voice Transcribe

**Judgment:** A focused, practical CLI skill for English speech-to-text on local audio files, well-suited for power users who live in the terminal and already use OpenAI’s APIs. **What it does well:** - Quickly transcribes a wide range of audio formats using an OpenAI transcription model. - Integrates cleanly into shell workflows (`transcribe <file>`, pipe to other tools, etc.). - Offers two mechanisms to improve accuracy over time: a vocabulary hints file (`vocab.txt`) and deterministic text replacements (`replacements.txt`). - Caches by audio file hash to avoid paying and waiting for repeated transcriptions of the same content. **Key risks / limitations:** - **Privacy & data handling:** All audio is sent to OpenAI’s servers; not appropriate for highly sensitive or regulated audio unless that’s acceptable under your data policies. - **Environment assumptions:** Requires `uv` and a working OpenAI API key configured in a `.env` file; the example path is user-specific and may confuse less-experienced users. - **Language constraint:** Assumes English; there’s no language detection or explicit multilingual handling. - **Network/API dependency:** Reliability and latency depend on OpenAI’s API availability and your network; there’s no mention of robust error handling or retries. **Recommended scenarios:** - You frequently receive short audio messages (e.g., WhatsApp voice notes) and want fast, scriptable transcription on your machine. - You’re comfortable with CLI tooling and want to chain transcription into broader automation (copy to clipboard, search, summarize, etc.). - You need to iteratively improve transcription for specific jargon, product names, or personal names via custom vocab and replacements. **Less ideal for:** - Non-technical users who aren’t comfortable with terminal tools or environment configuration. - Use cases that demand strict on-device processing or strong guarantees that audio never leaves your infrastructure. - Multilingual or language-detection-heavy workflows.