Downloads · 30 days
0
johnson-yo/termux_os-asset-sensevoice-htp-onnx
termux_os-asset-sensevoice-htp-onnx is a automatic speech recognition model from johnson-yo. Use it when you need speech turned into text. It is set up for onnx. The card lists the license as other.
This repository contains the portable ONNX source graph and the two small front-end files used by SenseVoiceSmall in Termux-OS. It is a raw Model Asset Package, not a service or an executable release.
Downloads · 30 days
0
Access
Public
Updated Sep 25, 2026
Repo size
4.8 GB
Likes
0
Public
Click a slice to open those files.
.onnx1.9 GB · 65%
From the Hugging Face model README
This repository contains the portable ONNX source graph and the two small front-end files used by SenseVoiceSmall in Termux-OS. It is a raw Model Asset Package, not a service or an executable release.
| File | Size | Role |
|---|---|---|
graph/generic/model.onnx | 937,255,762 B | fp16-safe source graph |
am.mvn | 11,203 B | acoustic mean/variance |
tokens.json | 352,064 B | CTC vocabulary |
ctx/t267/<target>/ | ~505 MB per target | fixed 16 s (T=267) HTP context for one device target |
ctx/t267/<target>/ holds one host-built HTP context per device target (for example
android-arm64-v73-qnn249, android-arm64-v79-qnn249): model.onnx (EPContext wrapper),
model.bin, am.mvn, tokens.json, and asset.json (capability descriptor with provenance and
per-file SHA-256). The package declares this asset once with target: "device"; the Registry lists
each target's files, and a device downloads only its own. Adding a target is an upload here plus
Registry rows — no package release.
The source graph is the Clip6e4 revision. It is the portable graph used when the device prepares its own execution artifact. The graph is intentionally not accompanied by a precompiled device-specific context: such artifacts are bound to the device architecture, runtime version, and model hash.
| File | Immutable source |
|---|---|
model.onnx | this repository at 33ad3cc1efcc036c9d6dce9073a37c06523523a4 |
am.mvn | FunAudioLLM/SenseVoiceSmall at 3847d57b6bdf2dd8875cb1508d2af43d80a16bf7 |
tokens.json | kautism/SenseVoiceSmall-onnx at 45311f13a91b52d4fceb7aefcfff87bb667b9c21 |
The Registry records each file's source, revision, remote path, exact size, and SHA-256. All three files are presented as one SenseVoice package root; the source repository is not inferred from the package id.
Framework installs and verifies raw bytes. The App/consumer may compile a device-local artifact into its own cache and owns loading, readiness, and cleanup. This repository never publishes or removes consumer cache data.
See NOTICE.md for provenance and licensing notes.