How to use Audio to Text
- Choose an audio or video file, record with your microphone, or click the example to see it work.
- Pick the language, or let it be detected, and a model: Fast, More accurate for Arabic and noisy recordings, or Best for Urdu on a computer.
- The first time, the speech model downloads once. Then the text appears as the recording is transcribed.
- Click a time to hear that part, then copy the text or download it as TXT, or as SRT or VTT subtitles.
Why use Audio to Text
- Private by design
- Whisper runs in your browser, so meetings, interviews and voice notes never leave your device. No sign-up and no minutes limit.
- 99 languages
- English, Arabic, Urdu, Hindi, French, Spanish and 93 more, detected for you. It can also translate any of them into English.
- SRT subtitle generator
- Download SRT or VTT subtitles with times, ready for YouTube, CapCut, Premiere or VLC.
- Record or upload
- Record a meeting or a lecture here, or choose MP3, M4A, WAV, OGG, WhatsApp voice notes, MP4 and other videos.
- Offline after the first download
- After the first download the speech model is kept in this browser, so transcription keeps working without a connection while the page is open.
- Meeting summary
- In Chrome with its built-in AI, get the main points of a meeting, written on your device too.
About this tool
Audio to Text is a free speech-to-text tool that runs Whisper inside your browser, on a phone or a computer. There is no sign-up and no app to install, and the recording is never uploaded. Whisper is OpenAI's speech recognition model, and you choose which size of it runs. Fast (Whisper base) suits clear speech and phones. More accurate (Whisper small) is much better with accents, background noise and Arabic. Best (Whisper large-v3-turbo, computers only) is the one for Urdu: on our test of 80 Urdu recordings it got 85% of words right, against 74% for More accurate, but it is a large download and slower. On a recent computer, Fast usually needs less time than the recording lasts; a phone is slower, so keep the page open while it works.
Whisper is very good with English and good with Arabic, but less accurate with Urdu, so check names, numbers and technical words before you use the text. For the best results, record one speaker close to the microphone with little noise around, and choose the language yourself for short clips.
Your file is decoded by your own browser and transcribed on your device: nothing is uploaded, and there is no account or limit. Only the speech model is downloaded, once, from this site.
Sources
- Whisper on GitHub (opens in a new tab): OpenAI’s model sizes and its note that accuracy varies by language
- Whisper large-v3-turbo model card (opens in a new tab)
- Robust Speech Recognition via Large-Scale Weak Supervision (opens in a new tab): the Whisper paper, arXiv 2212.04356
- Whisper’s licence (opens in a new tab): MIT
Frequently asked questions
Is this audio to text converter free, with no sign-up?
Yes. There is no sign-up and no minutes limit, because the transcription runs on your device, not on our servers. Long recordings simply take longer; on phones, recordings over an hour may run out of memory.
Is my recording uploaded?
No. The file is read and transcribed in your browser. Only the speech model is downloaded, once, from this site.
Which model should I choose?
Fast (68 MB) for clear English and on phones. More accurate (184 MB) for Arabic, strong accents, several speakers or background noise. Best (623 MB, computers with 8 GB of memory or more) for Urdu: it makes about half the mistakes of More accurate, but is the slowest.
How accurate is it with Urdu and Arabic?
Arabic is good with the More accurate model. For Urdu, use Best on a computer: on our test of 80 Urdu recordings it wrote 95% of the letters right (More accurate: 90%). Still read the text through before you use it.
Why did my Urdu come out in Hindi script?
Spoken Urdu and Hindi sound the same, so a speech model can write either. This page picks the script from your device (your last choice here, your browser's languages or your time zone) and keeps the whole transcript in that one script. If it is the wrong one, tap "Write it in Urdu" (or Hindi) above the transcript.
Can it make subtitles?
Yes. Download SRT or VTT subtitles with a time for each line, then upload them to YouTube or open them in your video editor or player.
Can it translate?
Tick "Translate it into English" and speech in any of its languages is written out in English.
Which files work?
Anything your browser can play: MP3, M4A, AAC, WAV, OGG and OPUS (WhatsApp voice notes), FLAC, WebM, and the sound of MP4, MOV and other videos.
Can it tell the speakers apart?
No. The transcript is one text, with the time of each line, so add the speaker names by hand afterwards.
Audio ko text mein kaise convert karen? (How do I turn a recording into text on my phone?)
Recording ko text banaye: audio ya video file chunein, model chunein aur transcribe hone dein; phone me bhi yahi tarika hai. Choose the voice note, recording or video on your phone or computer and the speech is written out, with a time for each line. It is free online with no sign-up and no minute limit, and the recording is never uploaded.
Does it work on iPhone and Android?
Yes. To transcribe or convert audio to text on an iPhone, iPad or Android phone, record a voice note or choose a file in Safari, Chrome or any modern browser, with no app to install. Fast (68 MB) and More accurate (184 MB) download once and then run on the phone itself; Best is kept for computers with 8 GB of memory or more. Keep the screen on while a long recording is transcribed, and on a phone keep recordings under about an hour.
Built and maintained by Saad Naeem
AI engineer, Doha · Updated