ihateaudio

Audio Transcriber

Speech in, text out, nothing uploaded.

Drop a video or audio file here

or paste one from your clipboard

MP4 · MOV · MKV · WEBM · MP3 · WAV and moreYour file stays on your device. Always.

How to use it

  1. Drop the recording in. The first run fetches Whisper, about 41 MB, once — after that this works with no internet at all.
  2. Press Transcribe. The model listens in thirty-second windows with a little overlap, so words are not cut in half at the joins.
  3. Read it, and fix what needs fixing. Names, jargon and acronyms are where every speech recogniser struggles, and it is faster to correct four words than to retype four hundred.
  4. Copy it to the clipboard or download it as a text file. If you want timings rather than prose, the subtitle generator produces the same transcript as SRT.

What the model is actually doing

Whisper does not recognise words the way older systems did, by matching sounds against a dictionary of pronunciations. It was trained on a very large quantity of audio paired with its transcript, and what it learned is closer to translation: given this stretch of sound, what text most probably accompanies it. That is why it punctuates, capitalises and spells out numbers without being told to — those were in the transcripts it learned from.

It also explains the model's characteristic strength and its characteristic failure. Because it is predicting probable text rather than decoding phonemes, it copes remarkably well with accents, mumbling and background noise, filling in from context the way a listener does. And because it is predicting probable text, when there is nothing to go on it will still predict something. Silence is the dangerous input, not noise.

Getting a better transcript

Clean the audio before transcribing, not after. Anything that makes speech easier for a person to follow makes it easier for the model too: the noise remover for a hissy or roomy recording, the silence remover to delete the passages where nobody is talking, and the volume booster if the recording is so quiet that the speech sits near the noise floor.

If the recording has music under the speech, running the vocal remover first sounds like a strange idea and works surprisingly well, because it is built to separate a voice from an arrangement.

Do not bother converting to a higher bitrate or a higher sample rate first. The model works at 16 kHz mono and the audio is resampled down to that regardless, so anything above it is discarded before the model sees it. A pristine 96 kHz file and a decent MP3 produce the same transcript.

What to check before trusting it

Proper nouns first. Names of people, places, companies and products are where errors concentrate, because the model is choosing the most probable spelling of a sound and an unusual name is by definition not probable.

Numbers second, particularly where they matter. "Fifteen" and "fifty" are a single vowel apart and both are plausible, so a figure in a transcript is worth confirming against the audio if anything depends on it.

And read any passage that looks unusually fluent for its position in the recording. A neat, well-formed sentence appearing where you remember a long pause is the signature of the model filling a gap rather than reporting one.

Questions

How accurate is it?
On clear speech in a quiet room, close to a careful human transcriber for ordinary words. On a phone recording in a cafe, noticeably worse. This is the small English Whisper model, chosen because at 41 MB it can sit behind a one-time download rather than being the reason nobody uses the tool. The larger models are better at accents, crosstalk and proper nouns, and they are several hundred megabytes, which is a different product.
Is my recording uploaded?
No. The only network request is the model coming down to you, and it is served from this domain rather than from Hugging Face or any other third party. Nothing about the audio, its contents or even the fact that you opened this page reaches anyone. That is the reason to use this for a medical appointment, a legal call, an interview under embargo or anything else you would not paste into a website.
What languages does it handle?
English only. This is the English-specific model, which is meaningfully more accurate at English than the multilingual one of the same size, and multilingual would be a second download to no benefit for most people here. If you need another language, the tool will produce something for it and that something will be wrong.
It invented a sentence that nobody said.
That is a known and slightly unsettling behaviour called hallucination, and it happens most in silence. Whisper was trained on speech, so when it is handed a passage with no speech in it — a long pause, background hum, music — it sometimes produces the most probable thing a person might have said instead. Running the silence remover over the recording first genuinely helps, because it deletes the passages where there is nothing to transcribe.
Two people are talking and it runs their words together.
Whisper transcribes speech; it does not identify who is speaking. Separating a recording by speaker is a different task called diarisation and needs a different model. What you get here is an accurate account of what was said with no indication of by whom, so for an interview you will be adding the speaker labels yourself.
How long can the recording be?
There is no fixed limit, and it is chunked so memory stays flat regardless of length. Time is the real constraint: transcription runs several times faster than real time, so an hour of audio is a matter of minutes rather than an hour. Long recordings are the case this handles best relative to paid services, since there is no per-minute charge and no upload.
Can I transcribe a video?
Yes, drop the video file straight in and the audio is extracted from it. If the file is very large it is quicker to run it through the video to audio tool first and transcribe the result, since that avoids decoding the video twice.