Skip to content

OmniSenseVoice

Fast speech recognition with precise timestamps

Visit official site

What OmniSenseVoice is

The model is described as a speech recognition model optimised from SenseVoice, designed for fast inference and precise timestamps. It is presented as a smarter and faster way to transcribe audio.

What you can do with it

  • Detect a language automatically or set one explicitly
  • Choose whether to apply inverse text normalisation
  • Run the model on a selected GPU or the default CPU
  • Use quantised models to speed up processing

Application scenarios

  • Transcribe live meetings into records with timestamps
  • Turn online course audio into timestamped notes for students
  • Feed real-time translation applications with fast speech recognition

Getting started

  • Install the speech recognition model
  • Set the language parameter, for example --language zh
  • Set the text normalisation option, for example --textnorm woitn
  • Specify the running device ID, for example --device-id 0

Pros & cons

✓ What we like

  • A GPU can be selected for running the model instead of the default CPU
  • A benchmark function is provided to evaluate model performance

! What to watch out for

  • The default running device is the CPU
  • Language and text normalisation options must be set through parameters

FAQ

Which languages are listed?

The material lists auto-detection, Chinese, English, Cantonese, Japanese and Korean.

Can it run on a GPU?

A specific GPU can be selected, while the default device is the CPU.

Last reviewed: 2026-11-15

More AI audio tools tools

View all →

How we review