Text can be searched, quoted, skimmed and pasted; a recording can only be listened to at the speed it was spoken. That difference is the whole reason people turn audio into text. A journalist needs the exact sentence from a forty-minute interview. A student wants lecture notes without replaying the lecture. A manager needs to find what was agreed in last week's call, and a podcaster wants show notes and a page search engines can read.
Typing it out by hand takes roughly four hours for every hour of speech, and it is tiring work. An automatic transcript arrives in minutes and gets most recordings right on the first pass, leaving you to fix a few names and technical terms instead of typing every word.
The transcript you get here is divided into short segments, each with its start time. That makes it easy to go back to the audio at the right moment when a word looks wrong, and it is the same structure subtitles are built from.
Nearly anything a phone, recorder or meeting tool produces: MP3 from podcast feeds and dictaphones, M4A from iPhone Voice Memos, WAV and FLAC from studio equipment, OGG and Opus from WhatsApp and Telegram voice messages, AAC, WMA and AMR from older devices. Video files are accepted too, and only their sound is used.
The file format has little effect on accuracy. What matters is how the recording was made. A voice close to the microphone in a quiet room transcribes almost perfectly, even at a low bitrate. A phone lying in the middle of a conference table, an echoing hall, music under the speech or people talking at once all cost accuracy, and no converter fully repairs that afterwards.
More than fifty languages are supported, and you do not have to say which one the recording is in. A file that switches between two languages is handled as well, although the less frequent language tends to come out a little rougher.
An hour of audio is usually ready in three to five minutes. You can leave the page open in the background; the progress is shown, and the result stays under the same address when you return.
Without an account you can convert 10 recordings, each up to 3 hours long, with 5 hours in total. For a long recording the page shows the opening part of the transcript as a preview, and the complete text is waiting in a free account, which takes a minute to create and needs no card. Signing up does not start a new allowance: guest and account share the same 5 hours.
The speech models run on Transcrify's own hardware. Your audio is not passed to an outside speech-recognition service, which matters when the recording is a client conversation, a medical consultation or anything else you would not hand to a third party.
Recordings uploaded without an account are temporary: the audio is removed after a day, while the transcript stays available so that you can still claim it by creating an account from the same browser.
The free allowance covers plain transcription. Paid plans add more hours and the tools that work on a finished transcript: automatic summaries, an AI chat that answers questions about the recording, subtitle export in SRT, VTT and ASS, translation, and export to PDF or Word. If one or two recordings are all you need, the free preview and a free account may well be enough.