Specifications
Size
100 MB - 300 MB
Architecture
Optimized Transformer
Latency
Very Low (<100ms)
Language
English
Developer / Creator
Useful Sensors
Download Source
Verified Repository Source
Hugging Face / Sherpa ONNX Model Registry
Open Model Repository (UsefulSensors/moonshine)Model Overview
Moonshine is a highly optimized, low-latency speech recognition model designed for real-time dictation on resource-constrained devices. It achieves similar accuracy to Whisper models while processing audio significantly faster with a smaller memory footprint. It runs locally in tapWhisper via Sherpa ONNX runtime.
Available Model Variants
| Model Name | File Size | RAM Usage | Format/Quant | Languages | Description |
|---|---|---|---|---|---|
| Moonshine STT (Tiny) | 103 MB | 300 MB | INT8 (ONNX) | English | Extremely fast Moonshine ONNX model for real-time English speech. |
| Moonshine STT (Base) | 239 MB | 650 MB | INT8 (ONNX) | English | Larger, higher-accuracy Moonshine ONNX model for English transcription. |