parakeet-local-asr

What This Skill Does

The parakeet-local-asr skill provides a robust, deterministic environment for running high-performance NVIDIA Parakeet ASR models directly on your local machine. By deploying this skill, you enable OpenClaw to process audio transcriptions entirely offline, ensuring maximum privacy and eliminating dependence on external cloud-based speech-to-text providers. The skill exposes an OpenAI-compatible API at http://localhost:9001, making it highly interoperable with existing transcription workflows and OpenClaw voice components. Whether you are using Ubuntu, Linux, or macOS (Intel or Apple Silicon), this skill manages the lifecycle of your local ASR server, from bootstrap and dependency installation to health monitoring and smoke testing.

Installation

Installation is streamlined through the OpenClaw hub. Run the following command in your terminal: clawhub install openclaw/skills/skills/hantok/parakeet-local-asr

Once installed, navigate to the source directory (default: ~/parakeet-asr) and follow the standard workflow:

1. Execute bash scripts/bootstrap.sh to install necessary dependencies.
2. Run bash scripts/start.sh to initialize the ASR service.
3. Verify operational readiness via bash scripts/healthcheck.sh.

You can override the default installation directory by exporting the PARAKEET_DIR environment variable before running the bootstrap script. Always ensure your environment meets the requirements for local model execution.

Use Cases

This skill is ideal for users who require strict data sovereignty, low-latency audio processing, or offline speech-to-text capabilities. It is particularly effective for developers building voice-controlled agents, privacy-conscious professionals transcribing sensitive meetings, or researchers experimenting with NVIDIA’s state-of-the-art Parakeet ASR architecture without the recurring costs of cloud API usage. It also serves as a critical fallback mechanism; if high-fidelity transcription is required, Parakeet can be configured alongside Whisper, balancing local privacy with linguistic accuracy.

Example Prompts

1. "OpenClaw, please install and configure the local Parakeet ASR service so I can transcribe my audio files offline."
2. "Run a smoke test on /home/user/downloads/interview.mp3 using the parakeet-local-asr skill."
3. "Can you check the health status of my local transcription server and ensure it is responding on port 9001?"

Tips & Limitations

Local ASR performance is heavily dependent on your hardware—specifically your GPU or Neural Engine capabilities. For intensive transcription tasks, ensure your machine has sufficient RAM and compute headroom. If you experience performance degradation, consider scaling your hardware or adjusting your model precision settings. Always prioritize health checks before assuming the service is ready for production. While the system supports Whisper as a fallback, remember that mixing models may result in varying transcription styles. Keep your local installation directory updated regularly to receive performance patches and compatibility improvements from the source repository.