Realtime Language Detection Demo

How much audio does a spoken language identification model need? Every model below runs in your browser and gives its answer after 20 ms, 40 ms, ... up to 2 s of audio.

1. Audio

or an example:
No audio yet.

2. Models

Each model downloads only when selected and you press Run. Files are cached by your browser.

Each model then chooses only among these languages (its probabilities are renormalized).

Marks each answer as correct or wrong.

Start of speech is found with Silero VAD.

Benchmark

Definitions

Duration
How much audio the model saw, counted from the start of the recording or from the start of speech.
Confidence
The model's probability for its top language (after restricting to your ASR languages, if set).
Stable decision
The shortest duration from which the model's answer no longer changes up to 2 s.

Prepared by

Kawshik Kumar Paul
Senior Software Engineer | Researcher
Department of Computer Science and Engineering (CSE)
Bangladesh University of Engineering and Technology (BUET)
Email: kawshikbuet17@gmail.com