Transcribe Audio to Text

Upload a recording and get the words back as plain text, plus SRT and VTT subtitle files. The file goes to our server, where OpenAI's Whisper speech model writes the transcript.

Drop an audio file to transcribe

Drag file here or pick from your device

MP3, WAV, FLAC, M4A, AAC, OGG, OPUS, AIFF, AIF, WMA · up to 200 MB · max 90 minutes

How to transcribe audio to text

Start with a sound file, such as a voice memo, a podcast episode or a recorded call. The file is uploaded to our server, so keep the tab open until you've downloaded what you need.

1
Pick your audio file
Drop the file on the box or click Browse file. A video such as an MP4 is turned away with a yellow note, so pull the sound out of the video first and upload that instead.
2
Choose the language, then click Transcribe audio
Leave Language on Auto-detect in most cases, since the model guesses the language from the first 30 seconds, so if the recording opens with music or with someone speaking another language, pick the right one from the list before you click Transcribe audio.
3
Check the words and download
When Transcription complete appears, the text is on the TXT tab with Copy to clipboard under it, and the SRT and VTT tabs hold the subtitles. Click Download all to save all four files at once. The links stop working after 30 minutes, so download before you step away.

Three captures of the transcribe tool: an MP4 turned away at upload, a 90 second M4A set to Auto-detect, and the finished job on the SRT tab

Guide

Audio to text and SRT subtitles: what you get back

Every run returns the same four files, whatever you plan to do with them. TXT is the plain words in one long paragraph, with no times and no speaker names. SRT and VTT hold the same words cut into short lines, each with the moment it starts and stops. JSON is the full answer from the speech model, which matters mostly to people writing their own code.

The waveform of a 90 second test clip above its 14 subtitle lines, marking a 17 second line and a 0.6 second line, plus the two phrases it misheard

We ran a 90 second reading of the Gettysburg Address and a 48 second Spanish fable through the tool. The results were good enough to work from, but not good enough to publish without a read-through.

Which AI writes the transcript

The words come from Whisper, OpenAI's speech-to-text model, which we run through a service called Replicate; Replicate lists the version as large-v3. Your file goes from our server to Replicate, and Replicate deletes its copy after an hour.

How accurate it was

Out of about 150 words in the Gettysburg test, two phrases were wrong. The title came back in capitals as GEDDESBURG, and "so dedicated, can long endure" came back as "so done, can be a good nation". The rest of Lincoln's words were exactly right, and the date came out as "November 19, 1863". Names and odd phrases are where to look first, so read the text against the recording before you quote it.

Using the SRT file for subtitles

SRT is the file to upload to YouTube or open in a video editor, while VTT is for players built into web pages. The lines follow the speaker's pauses, not a set length, so they can run long; one of ours had 96 characters, which is two lines on most screens. Timing needs a check too, because our first line stayed up for 17 seconds while showing only the title, and line 6 was on screen for just 0.6 seconds. Fix those in your video editor before the video goes out.

Length and size limits

The most audio the tool takes is 90 minutes, and that check runs after the upload, so a longer file uploads in full and only then fails. The box says up to 200 MB, but the network service in front of our server turns away anything over about 100 MB with "File is too large". An MP3 uses about 1 MB a minute, so 90 minutes fits, but a CD-quality WAV uses about 10 MB a minute and passes 100 MB at around 10 minutes. Turn a long WAV into MP3 first with WAV to MP3; the words come out the same.

Videos, voice notes and WebM files

Only sound files are accepted: MP3, M4A, WAV, FLAC, OGG, OPUS, AAC, AIFF and WMA. For a video, get the sound out first with Extract Audio and upload the MP3 it gives you. A WebM recording is also turned away; WebM to MP3 fixes that, and our Voice Recorder can save a new recording as MP3 to begin with. iPhone voice memos are M4A files, which work as they are; our English test clip was an M4A.

How long your files are kept

Your audio is deleted from our server as soon as the transcript is done, but if the job fails, the audio waits for the nightly cleanup, which removes anything older than an hour. The four transcript files stay until that same cleanup, but their download links stop working 30 minutes after they're made. Cancel only stops this page from checking on the job, so the server still finishes it.

Last updated: 22 October 2026 · ChangeMyFile

Tested: 22 October 2026, headless Chrome 156 on a Mac against the live server, two jobs. Job 1: the first 90 s of the LibriVox public domain reading of the Gettysburg Address, saved as an M4A (AAC 96 kbps mono, 1.1 MB), Auto-detect. Done 20.2 s after clicking Transcribe audio; Detected: English, Duration 1 min 30 sec, model openai/whisper. TXT 828 bytes and 150 words in one paragraph, SRT 14 cues (1,294 bytes), VTT 1,269 bytes, JSON 8,906 bytes with per-segment start and end times. Two misheard phrases, listed above; cue 1 ran 0 to 17 s, cue 6 lasted 0.6 s, longest cue 96 characters. Job 2: LibriVox public domain reading of Aesop's "El Cuervo Enfermo" in Spanish (MP3, 48 s, 377 KB) with Translate to English ticked. Done in 26.3 s, Detected: Spanish; TXT, SRT (8 cues) and VTT in Spanish, English only in the JSON "translation" field without timings. An MP4 and a WebM were refused in the browser before upload. Upload size probed with non-audio files: 99 MB reached our server, 101 MB and 150 MB were refused with 413 by Cloudflare. All download links returned the files straight after each job.

FAQ

Audio transcription questions

At a glance

Where it runs: On our server, with the speech model at Replicate
Speech model: OpenAI Whisper, via Replicate
Files you get: TXT, SRT, VTT and JSON
Longest audio: 90 minutes
Size limit: About 100 MB
Languages: Auto-detect, or pick one
Download link: Works for 30 minutes
Files deleted: Your upload once the transcript is done, the rest in a nightly cleanup
Account: Not needed

Can't find your answer? Visit our Help center.

Is this audio to text tool free?

Yes, it's free, and you don't need to sign up. Each file can be up to 90 minutes long and about 100 MB.

Can I turn an MP3 into an SRT subtitle file?

Yes, every job makes an SRT file along with the text, so open the SRT tab after the job finishes and click Download .srt. Read through the timing before you use it, since a line can sit on screen far too long or flash past before anyone can read it.

How long does it take to transcribe audio?

Our 90 second clip took 20 seconds from the click to the finished text, and a 48 second clip with translation took 26 seconds. Longer files take longer, and the page says to allow several minutes for long ones.

How accurate is the transcript?

In our English test, two phrases out of about 150 words were wrong, and one of them was a place name. A clear recording of one person talking gave us good results, but plan to correct names, numbers and anything said over music or background noise.

Can it transcribe other languages?

Yes, it picked out Spanish on its own in our test and wrote the Spanish words correctly. The menu names 12 common languages, and choosing Other lets you type a two or three letter language code instead, such as nl for Dutch.

Can it translate audio into English text?

Not in the text and subtitle files yet: with Translate to English ticked, our Spanish test still came back in Spanish, and the English version was only in the JSON download, as one block of text without times.

Can I transcribe a video?

Not directly, because the tool accepts sound files only, so take the sound out of the video first with our Extract Audio tool, then upload the MP3 file it gives you here.

Does it label who is speaking?

No, the text and subtitles carry no speaker names, so for an interview or a meeting you'll need to add them yourself while you read through the transcript.

Are my recordings private?

Your file is uploaded to our server and passed to Replicate, the service that runs the speech model. We delete the audio when the transcript is done, Replicate deletes its copy after an hour, and the transcripts go in our nightly cleanup.

Why does it say my file is too large?

Uploads over about 100 MB are turned away before they reach our server, even though the box says 200 MB. Saving the recording as an MP3 first usually brings it well under that.

Audio Cutter Merge Audio Compress Audio Voice Recorder