STT Bench — Transcription comparator

— No model selected —

00:00.0

— idle —

Prompt for multimodal models (Voxtral Small, Gemini)

Settings

API Keys

Provider

Stored locally in your browser (localStorage) and offered to Chrome Password Manager for cross-device sync. Never sent anywhere except to the corresponding APIs.

Models

Tick the models you want to use. Tip: select a single model for the cleaner one-up layout and the K shortcut to copy.

Default prompt (multimodal)

Used by Voxtral Small and Gemini 3.1 Flash Lite (which need a text prompt alongside the audio).

Silence trimming (Silero VAD)

Trim silences (start, middle, end) before sending

Runs a small Silero VAD model locally in your browser to remove silent passages from the audio. Reduces Whisper hallucinations and saves API cost. Adds ~200–400ms of processing per recording (after first use; the model is cached).

Mono-model flow

Auto-transcribe and copy when recording stops

Only applies when exactly one model is selected. Skips Send and copies the result to clipboard automatically.

Keyboard shortcuts

Toggle recording Press to start/stop recording

Cancel recording Discards the current recording without sending it

Send / Transcribe Sends the audio after recording

Copy result (mono-model only) Copies the single result to clipboard

Open settings Opens this drawer

Click an input then press the desired key. Shortcuts are disabled when typing in a text field.