The short version: drop your audio (or video — it'll pull the audio track) into a speech-to-text tool, pick the spoken language instead of "auto" if you know it, and export as SRT if you need timestamps or TXT if you just want the words.

Here's how to get a clean result and avoid the common mistakes.

Get the language right

Auto-detect is convenient but it's the #1 cause of garbled transcripts. If the recording is mostly English, pick English. For mixed audio (e.g. English with some Chinese terms), choosing the dominant language usually beats auto-detect. It tells the model what to expect and cuts the weird substitutions.

Clean audio in, clean text out

Transcription quality tracks audio quality:

  • One clear speaker close to the mic transcribes near-perfectly.
  • Crosstalk, echo, background music drops accuracy fast. Nothing fixes a bad recording after the fact, so capture well: quiet room, mic close, one person at a time.
  • Long silences and filler are fine — just expect "um"s in the output you'll trim.

Timestamps and subtitles

If you're subtitling a video or need to jump to a moment, export SRT — each line carries a start/end time. For notes or a searchable document, TXT is enough. Most tools (ours included) give sentence-level timestamps you can click to jump to that point in the audio.

On-device vs cloud — which to use

  • On-device (in-browser): the audio never leaves your computer. Best for anything sensitive — interviews, medical, legal, internal meetings. The trade-off is speed: it depends on your hardware and can be slow on older machines.
  • Cloud: faster and often more accurate on long or noisy files, but the audio is sent to a service for processing.

Rule of thumb: sensitive or short → on-device; long, noisy, or you're in a hurry → cloud.

Quick workflow

  1. Open Speech to Text and add your audio or video.
  2. Set the language (not "auto" if you know it).
  3. Transcribe, then read through once — fix names and jargon the model guessed.
  4. Export SRT (for subtitles) or TXT (for notes).

A pass of human cleanup on names, numbers and technical terms takes a few minutes and makes the transcript actually usable. No tool gets those 100% right.

Have a recording to transcribe? Drop it into Speech to Text and export your transcript.