FAIND – Flash AI News DailyFAIND – Flash AI News Daily ← Back to the news

FAIND app · free tool

Audio and video to text, or extract the audio: free, on your own device

Free. Runs in your browser. Your recording never leaves your device. Check it yourself ↓

Drop a file here, or tap to choose MP3, WAV, M4A, MP4, MOV, WebM and more · up to 400 MB

Extract the audio from a video

Need only the sound of a video? Drop the file in, choose Extract the audio, pick a format and download it. The same works on a recording you want in another format, for example an M4A voice memo as MP3. Your browser decodes the sound and writes the new file, so the video never leaves your device.

FormatBest forSize per minute
MP3Sharing and playing anywhere. Compressed, a little detail is lost.about 1 MB (Original), 0.5 MB (Voice)
M4A (AAC)Small files, good quality, Apple devices. Only in browsers that can encode AAC.about 1 MB (Original), 0.4 MB (Voice)
WAVEditing and archiving. Uncompressed, nothing more is lost after decoding.about 10 MB (Original), 2 MB (Voice)

Two quality settings: Original keeps stereo at 44.1 kHz (up to 30 minutes), Voice is mono at 16 kHz, much smaller and enough for speech (up to 60 minutes). Longer sound needs a lot of memory, so split very long files first. MP3 is encoded with LAME through lamejs, a JavaScript port released under the LGPL (license), served by FAIND as a separate, unmodified file. The sizes are estimates from the bitrates.

Subtitles that follow the translation

Besides plain text you get SRT and VTT subtitle files, and they follow the translation. Transcribe once, pick another language under the result, and download the subtitles in that language with the original timing. The translation also runs on your device.

How it works

The page loads Whisper, an open-weights speech recognition model from OpenAI, into your browser. On computers and phones that support WebGPU it uses the graphics chip; otherwise it falls back to the processor, which is slower. Nothing is sent to FAIND for processing. Running a trained model like this is what AI people call inference, and here it happens on your own device.

The first run downloads the model, which your browser then keeps. The download comes from Hugging Face and the code from jsDelivr, two public hosting services; they see a normal file request, never your recording. See the privacy notes.

Check it yourself: nothing is uploaded

You do not have to take our word for it. Any browser can show you where the data goes.

  1. Open your browser's developer tools (F12, or Cmd+Option+I on a Mac) and go to the Network tab.
  2. Choose a file and run the tool.
  3. Read the list of requests. On the first run you will see the page code and the speech model being downloaded; on later runs the model comes from your browser's own storage. No request carries your file: there is no upload (no large POST or PUT). The only other traffic is a cookie-free visit counter that sends the page address, never your file.

What people use it for

Interviews, lectures and meetings

Record on your phone, drop the file here and read it instead of replaying it. Search the text for the part you need, and use the timestamps in the SRT tab to find the exact moment in the audio.

Subtitles for a video you made

Transcribe the video, download the SRT or VTT, and upload it to your video platform or editor. If your audience speaks another language, translate first and download the subtitles again.

Voice notes and messages

Long voice messages are faster to skim as text. Save the message as a file, choose it here, and copy what you need.

A recording in another language

Choose the language of the speech, transcribe, then pick the language you want from the "Translate to" menu under the result. You get both texts, so you can compare them.

What to expect

OutputPlain text with punctuation, plus SRT and VTT subtitles with timestamps.
SpeakersNot separated. You get one continuous transcript.
SpeedDepends on your device. A recent laptop with WebGPU is quick; an older phone can be several times slower than the recording.
LimitsText: 60 minutes and 400 MB per file. Audio extraction: 30 minutes (Original) or 60 minutes (Voice), 400 MB. Keep the tab open while it works.
AccuracyBest on clear speech in widely spoken languages. Noise, accents and overlapping voices reduce it.

Questions people ask

How do I turn audio into text for free?

Drop the file into the box at the top of this page, choose the language (or leave it on automatic) and press Transcribe. When it finishes you can copy the text or download it. There is no account, no watermark and no minutes limit, because your own device does the work.

How do I turn a video into text?

Same steps: drop the video in (MP4, MOV, WebM and most other formats work). The page reads the sound track and transcribes it. You get plain text and, if you want, SRT or VTT subtitle files with timestamps. If your browser cannot read the file, convert it to MP4 or MP3 first.

Can I transcribe a voice message or a voice memo?

Yes. Save the voice message or memo as a file on your phone or computer and choose it here. iPhone voice memos are usually M4A, which works in most browsers. Some voice-message formats, such as WhatsApp's OPUS or OGG, may not open in Safari: in that case convert the file to MP3 or M4A first.

Can I choose the language of the recording?

Yes. Pick it from the Language menu (15 are listed) or leave automatic detection on. Choosing it yourself usually works better on short or noisy clips. Whisper itself handles about a hundred languages, but this page only lists the most common ones.

Can it translate the transcript?

Yes, into the same 15 languages, from the menu under the result. In recent desktop Chrome the translation uses the browser's built-in on-device translator, with nothing to download. Elsewhere it offers a free open model (several hundred MB, stored by your browser after the first use). Either way the text stays on your device. Machine translation can be wrong on names, idioms and technical terms.

Can I get subtitles (SRT or VTT) from a video?

Yes. After the transcription, open the SRT or VTT tab and download the file. Each line has start and end times, and the same works after translation, so you can also get subtitles in another language. Check the timing and line breaks before you publish: they follow the speech, not subtitle style rules.

Is my audio uploaded anywhere?

No. The speech model is downloaded to your browser and the transcription runs on your own device. The file is never sent to FAIND or to any other server.

Is it really free?

Yes: no account, no watermark, no minutes limit. The work is done by your own computer or phone, so it costs FAIND nothing to run.

How accurate is it?

Good on clear speech in widely spoken languages, weaker with background noise, strong accents, music or people talking over each other. Always read the result before you use it, especially names, numbers and technical terms. The Better quality option is slower but usually more accurate.

Does it tell speakers apart?

No. The output is one continuous transcript with timestamps. It does not label who is speaking.

Can I paste a YouTube or Zoom link?

No. The page works on files you choose from your device, not on links, because a browser cannot take the sound out of a video page. If you have the right to use a recording, save the file first and then choose it here.

Why is the first run slow?

The first time, your browser downloads the speech model. It is then stored by your browser, so later runs start straight away.

How long can the file be?

Up to 60 minutes and 400 MB per file. Long files take longer on slower devices; a computer with a recent browser is the best choice. For longer recordings, split the file first.

How can I check that my recording is not uploaded?

Open the developer tools of your browser (F12), go to the Network tab and run a transcription. You will see the speech model being downloaded the first time, and no request that sends your recording anywhere. The steps are in the "Check it yourself" section above.

Is it safe for confidential interviews or meetings?

The recording stays on your device and is not handed to any server for processing, which is why this approach suits sensitive audio. Two cautions: the transcript is also stored on your device, so protect it like the original, and the rules on recording other people depend on where you live, so ask for consent where it is required.

How is this different from an online transcription service?

Many online services upload your file to their servers and may ask for an account or limit the minutes. Here the work happens on your device: no upload, no account, no quota. The trade-offs are that speed depends on your device, speakers are not labelled and nobody reviews the text for you.

How do I translate an audio recording into English?

Choose the language of the speech (or leave automatic detection on) and press Transcribe. Then open "Translate to" under the result and choose English. You can read, copy or download the translation, including as SRT or VTT subtitles.

How do I extract the audio from a video for free?

Drop the video into the box at the top of this page and choose "Extract the audio". Pick MP3, WAV or M4A, then press Extract audio and download the file. It is free, needs no account and leaves no watermark, and the video is not uploaded: your browser does the work.

How do I convert an MP4 video to MP3?

Drop the MP4 file in, choose "Extract the audio", leave the format on MP3 and press Extract audio. You get an MP3 file with the sound of the video, which you can play in the page or download. MOV and WebM files work the same way, as long as your browser can play them.

Which audio formats can I get, and which one should I choose?

MP3 plays almost everywhere and is small, so it is the safe choice. M4A (AAC) is similar in size with slightly better quality at the same bitrate, and suits Apple devices; it appears only in browsers that can encode AAC. WAV is uncompressed, so it is large (about 10 MB per minute in stereo) but loses nothing after decoding: use it for editing.

Does extracting the audio lower the quality?

The page decodes the sound of your file and writes it again in the format you choose. WAV keeps the decoded sound intact. MP3 and M4A are compressed, so they lose a little detail, which is normal for these formats. It is not a bit-for-bit copy of the original audio track. The "Original" quality keeps stereo at 44.1 kHz; the "Voice" option is mono at 16 kHz, much smaller and enough for speech.

Is the video uploaded when I extract the audio?

No. The audio is decoded and encoded by your own browser. The file is never sent to FAIND or to any other server. You can check it in the Network tab of your browser's developer tools, as explained in the "Check it yourself" section.

Why can I not choose M4A in my browser?

M4A needs an AAC encoder built into the browser, and not every browser has one. Recent desktop Chrome and Edge usually do; others may not. When it is missing, the option is greyed out and you can use MP3 or WAV instead.

How long can the video be for audio extraction?

With the Original quality, up to 30 minutes, because stereo sound at 44.1 kHz takes a lot of memory while it is processed. With the Voice quality (mono, 16 kHz), up to 60 minutes. The file size limit is 400 MB in both cases. For longer videos, split the file first.

Can I extract the audio from a YouTube link?

No. The page works on files you choose from your device, not on links. If you have the right to use the video, save the file first and then choose it here.