Speech to Text Is Live — Transcripts, Not Captions

Speech to text is live today, and it does one thing: turn a recording into a plain transcript and a duration. It does not produce captions, subtitles, SRT or VTT files, timestamps, word-level timing, or speaker labels — just text.

If you came here expecting subtitles, this isn't that tool. This post explains exactly what you do get, why that's genuinely useful on its own, and where the numbers land: file limits, supported formats, and how it's billed.

Upload an audio or video file and you get back two things: a plain-text transcript of everything said, and the file's duration. That's the whole output. There are no timestamps next to individual words, no speaker A / speaker B labels, and no formatting for a caption track — just the words, as one continuous piece of text.

Every plan can transcribe, but the ceiling on a single file depends on your tier.

Anything past roughly 18 minutes runs as an asynchronous job rather than a spinner you sit and watch. You upload the file, leave the page, and come back once it's finished. Accepted formats cover most of what people actually have sitting around: mp3, wav, m4a, ogg, flac, webm, mp4, mov, mkv and avi — audio or video, either works.

30 languages are recognized. The language picker is a hint, not a requirement. Leave it on auto-detect and the transcript engine identifies the language itself, which is the right fallback for anything you're not sure about.

Transcription is priced by duration, not by characters, because there's no text going in to count. It costs 15 points per minute of audio: a 20-minute file costs 300 points, a full hour costs 900. That's a different rate from generation, which is priced from the script you paste, and it's worth keeping the two separate in your head.

If you've read anywhere on this site about SRT export or word-level subtitles, that's real, and it's worth being precise about which direction it runs. Those posts, including SRT export from text to speech, describe generating audio from a script: you supply the text, the engine returns word timing alongside the audio, and an SRT file is built from that timing.

Speech to text runs the opposite way. You supply a recording and get a transcript back — no timing, no caption file, nothing built from it. Captions come out of text going in, not out of audio going in.

Speech to text and text to speech run in opposite directions, but they connect in one useful way: once you've got a clean transcript, you can hand it back to a voice and generate narration from it — a recap, a spoken version of your own notes, a written interview turned back into audio. Here are six voices worth auditioning for that second step. Browse the full voice library for the rest.