🗣️ Vosk ASR to RKNN Conversion for RK3588

This repository provides scripts and instructions to convert Vosk speech recognition models (in ONNX format) into RKNN format, optimized for the RK3588 NPU.

Use cases include real-time offline transcription on devices such as Orange Pi 5 and Radxa Rock 5B.

---

🎯 Purpose

• Convert Vosk ONNX models to RKNN
• Enable low-latency ASR using the RK3588 NPU
• Avoid heavy CPU usage during audio transcription

---

🧱 Dependencies

• Python 3.8–3.10
• RKNN Toolkit 2
• onnx==1.14.1
• numpy
• Vosk ONNX models (encoder.onnx, decoder.onnx, joiner.onnx)

🔧 Install Python dependencies:

pip install onnx numpy