Model Hub & Registry
Curated lightweight on-device models with automatic downloading and SHA-256 integrity checks.
Whisper Models (GGML Quantized)
| Model Identifier |
Parameters |
Quant Level |
Disk Size |
Peak RAM |
Target Use Case |
whisper-tiny |
39M |
q5_1 / f16 |
~42 MB |
~150 MB |
Realtime voice assistant, fast command recognition |
whisper-base |
74M |
q5_1 |
~82 MB |
~250 MB |
General transcription (Default) |
whisper-small |
244M |
q5_1 |
~190 MB |
~500 MB |
High accuracy lectures, podcasts |
whisper-medium |
769M |
q5_1 |
~520 MB |
~1.5 GB |
Golden balance for professional meeting minutes |
Vosk Models (X-Vector & STT)
| Model Identifier |
Disk Size |
RAM Footprint |
Description |
vosk-model-small-ko-0.22 |
42 MB |
~100 MB |
Lightweight Korean STT acoustic & language model |
vosk-model-spk-0.4 |
14 MB |
~20 MB |
128-dimensional X-Vector speaker embedding extractor |
Sherpa-ONNX Models
| Model Identifier |
Disk Size |
RAM Footprint |
Description |
sherpa-zipformer-ko-2024 |
~150 MB |
~300 MB |
Ultra-low latency streaming Zipformer model |
sensevoice-small-onnx |
~200 MB |
~400 MB |
Non-autoregressive ultra-fast multi-lingual STT |
3dspeaker-campplus |
~25 MB |
~50 MB |
CAM++ speaker verification and embedding model |
CLI Model Management
# List installed models
termux-stt models list
# Download a specific model
termux-stt models download whisper-medium
# Remove a cached model
termux-stt models remove whisper-small
