Speech to Text: the guide

Turn audio into timestamped text, entirely in your browser - the audio never leaves your device. Entry point: https://lab.stsqz.com/transcribe.

Supported formats

Common audio formats all work (mp3, wav, m4a, ogg, and more). Files over five minutes are trimmed to the first five minutes by default; trim longer material first in the Audio & Video tool.

Choosing a model tier

All three tiers are Whisper-family models. You download once, the model caches in your browser, and it keeps working offline. Switching tiers downloads that tier's model; earlier caches stay put so you can switch back anytime.

What happens during transcription

The engine slices your audio into ~30-second chunks and recognizes them one by one while the page shows live progress. Results come back as segments with start and end timestamps. You can set the language manually - if the audio is Mandarin, say so, and the engine skips language guessing, which is both faster and more accurate.

Export

• Copy directly - one click copies the full text, ready to paste anywhere
• SRT subtitles - export a standard SRT file with the timeline already aligned; drop it straight into your editor or player

Pair it with the Audio & Video tool for a complete subtitle workflow: transcribe, export SRT, then burn the subtitles into the video.

FAQ

The results aren't accurate

• Step up to a higher tier
• Keep the source clean; heavy background music hurts accuracy
• Mandarin Chinese works best; for other languages, try the tiny tier first to confirm it's usable

Stuck on loading

First use downloads the model - a few hundred MB can take minutes on a slow connection. The page shows progress; don't close it. You only download once.