Downloads · 30 days
0
tubome/whisperapp-asr-models
whisperapp-asr-models is a machine learning model from tubome. Use it for the machine learning task on the model card, and read the license before you ship it in a product. The card lists the license as apache-2.0.
On-device ASR models for WhisperApp Android.
Downloads · 30 days
0
Access
Public
Updated Mar 11, 2026
Repo size
1.6 GB
Likes
0
Public
Click a slice to open those files.
.gz1.6 GB · 100%
From the Hugging Face model README
On-device ASR models for WhisperApp Android.
Architecture: Moonshine v1 (full-attention encoder), multilingual fine-tuned
Format: sherpa-onnx 2-file format (encoder_model.ort + decoder_model_merged.ort + tokens.txt)
Quantization: int8 (ONNX Runtime dynamic quantization)
Runtime API: ("v2" refers to the export format, not the model generation)
| Model | Size | Languages |
|---|---|---|
| moonshine-base-{lang} | ~107 MB | ja, en, zh, ko, ar, vi, uk |
| moonshine-tiny-{lang} | ~58 MB | ja, en, zh, ko, ar, vi, uk |
Architecture: OpenAI Whisper (99+ languages)
Format: sherpa-onnx ONNX format ({size}-encoder.int8.onnx + {size}-decoder.int8.onnx + {size}-tokens.txt)
Quantization: int8
Use case: Fallback for languages not covered by Moonshine Flavors
| Model | Size |
|---|---|
| whisper-small | ~222 MB |
| whisper-base | ~90 MB |
| whisper-tiny | ~58 MB |
Each contains a directory with the model files:
These models use the Moonshine Flavors (v1 architecture) with a full-attention encoder. This is distinct from Moonshine v2 (Streaming) which uses a sliding-window encoder and currently supports English only. If/when Moonshine v2 Streaming adds multilingual support, this repository may be updated accordingly.
Apache 2.0