●
Room App
· on-device Whisper → live captions
connecting
0 viewers
mobile viewer ↗
1 · Room
Room name
Create room
Room code
------
idle
End room
2 · Whisper model (runs in this browser)
Model
whisper-tiny.en · fastest (~40 MB)
whisper-base.en · balanced (~75 MB)
whisper-small.en · best, slow on CPU (~250 MB)
whisper-base · multilingual
Compute
auto (WebGPU if available, else WASM)
WebGPU
WASM (CPU)
Load model
Not loaded. Weights download once and are cached by the browser.
3 · Microphone
Start transcribing
Stop
Chunking settings
Cut on silence longer than (ms)
Minimum chunk (ms)
Maximum chunk (ms)
Speech threshold (RMS, auto-adapts above noise floor)
Send
Stats
0
chunks sent
–
avg inference
–
real-time factor
0
queue backlog
–
publish ack RTT
–
clock offset / ws rtt
Transcript
clear
Segments appear here as Whisper finishes each chunk. Each shows audio length, inference time and the server ack.