Audio to SRT: free accurate transcript now, honest timestamps next

What an SRT file actually is

SRT (SubRip Subtitle) is a plain-text format that video players and platforms read to show captions at the right moment. Every caption is a small block with three parts, and blocks are separated by a blank line.

• 1 · 00:00:01,000 --> 00:00:04,200 · Welcome back to the channel.
• 2 · 00:00:04,400 --> 00:00:07,900 · Today we are testing three budget microphones.
• 3 · 00:00:08,100 --> 00:00:11,600 · Starting with the cheapest one first.

Why audio to SRT needs real timestamps

Stack blocks like the three above and you have a complete subtitle file. Notice which part carries the weight: the timing lines. The words are the easy half. The millisecond start and end of every block is what makes it a subtitle file instead of a transcript — and it is the half you cannot invent.

It is tempting to think you can convert a transcript to SRT by splitting the text into chunks and spacing the chunks evenly across the runtime. Some converters do exactly that. The result drifts, because people do not speak at a constant rate. A speaker rushes one sentence, pauses before the next, stops for three seconds to think, then talks over a jingle. Timings guessed from text length — fifteen characters per second, say — separate from the voice within a minute or two, and every caption after that lands early or late.

The only trustworthy source of timing is measurement of the audio itself: the recognition pipeline reporting when each segment was actually heard. This site does not export timestamps until that measurement is verified end to end, because publishing guessed timing would be worse than publishing none.

Audio to SRT today: transcribe now, subtitle later

You can do the hard half — an accurate transcript — for free on this page right now, then add timing in a subtitle editor. The whole flow takes six steps.

For your own YouTube uploads there is an even shorter path that skips manual timing entirely — see the FAQ below.

1. Upload your audio in the transcriber above: MP3, WAV, M4A, AAC, OGG, or OPUS, up to 5 minutes and 20 MB per file on the free tier. Or record straight from the microphone.
2. Leave language detection on auto — 40 locales are covered — or pick the language if you already know it.
3. Choose Best accuracy mode. For anything you plan to caption and publish, there is no reason to trade precision for speed.
4. Run the transcription, then copy the text or download it as a .txt file. It arrives punctuated and capitalized, which matters for subtitles: caption blocks read as sentences, not as a lowercase word stream.
5. Open any subtitle editor that supports transcript import — most desktop and web subtitle tools have a paste-transcript or import-text option — and load your audio or video alongside the text.
6. Set the block timings while listening, then export the SRT from the editor. Because the words are already correct, you are only nudging boundaries, not typing.

What the transcriber above does — and does not

The transcription runs on NVIDIA's Nemotron 3.5 ASR, a 0.6B-parameter cache-aware FastConformer-RNNT streaming model. This site is an independent tool built on that model — not affiliated with, sponsored by, or endorsed by NVIDIA. Here is the free tier, spelled out.

• No SRT or VTT export yet — that gap is the entire reason this page exists.
• No word or segment timestamps in the output.
• No speaker labels; the transcript is one continuous text.
• No video files. Extract the audio track first, then upload that.
• Transcription runs after the upload or recording finishes — it is not a live caption overlay.
• New Pro purchases are paused. Existing Pro license keys still activate in up to 3 browsers.

SRT export is on the roadmap — without a fake date

SRT export ships when segment timestamps are verified end to end: the pipeline has to report segment timing, preserve it through upload and decoding, and produce exported files that stay in sync on real recordings — long ones, noisy ones — not just short clean clips.

No launch date is promised here, because a date would be a guess and this page does not do guesses. What can be said plainly: when SRT export arrives, it will be built on measured timing from the recognition pipeline, not on characters-per-second estimates. Until then, the transcript is the deliverable, and the transcribe-now-subtitle-later workflow above remains the honest way to get from audio to SRT.