Back to the marketplace · Model

SenseVoice local speech

Turn speech into composer text on the computer running Teloa. Supports Mandarin, Cantonese, English, Japanese and Korean.

General Communication & Collaboration

Copy this sentence and send it to any colleague in Teloa:
Find “SenseVoice local speech” (teloa.model.sensevoice) in the Models category of Teloa Market and open “Prepare local speech”.

Source: Teloa official
GitHub: Entry file: catalog/models/teloa.model.sensevoice.json
License: MIT
Review: Reviewed by Teloa on 2026-09-26
Compatibility: Needs configuration · Teloa >=0.2.0-alpha.7 · DSH 0.1.7-rc.1
Enable the native Voice Input extension, restart Teloa, then confirm preparation. Native targets include macOS arm64/x64, Linux glibc arm64/x64 and Windows x64; actual readiness is determined by the host.
Permissions and requirements: Speech recognition runs locally on the Teloa host.
Prepared and managed by the native DSH speech extension using SenseVoiceSmall ONNX and Silero VAD.
The default INT8 main weight is about 239 MB, plus tokens and VAD. File size is not runtime memory; the native preparer verifies versions, hashes and download progress.
Installation is on the Teloa host, not the device viewing this page. Initial preparation requires a download; recognition then runs locally. Disabling the extension retains cached files.
Weights follow the FunASR model license. The source-code MIT license does not replace the weight license; retain the required notices when using or redistributing the model.
FunASR Model Open Source License Agreement
Data flow: Initial preparation downloads weights from native sources. Speech recognition runs on the Teloa host without uploading audio to a model provider. This site collects no user information.

Entry teloa.model.sensevoice · version 1.0.0