Transcribe audio to text automatically

Audio-only recordings are a harder case than video

There is no picture to fall back on, which matters more than it sounds. A listener resolves an unclear word from a speaker's mouth or a slide behind them; a recogniser working from a phone recording of a meeting has only the waveform. Room echo, a microphone on a table between four people, and two speakers talking over each other are the three things that cost the most accuracy.

None of that is fatal — it is why the transcript is editable before you export it. But it is worth knowing which recordings will need a read-through and which will not.

What comes back, and in which format

A plain TXT transcript is usually what you want from audio: continuous text, no timecodes, ready to paste into notes or a document. SRT and VTT come from the same job if you need timings — for a podcast player that shows a synchronised transcript, or to line the text up against the original recording.

• TXT — continuous text with no timecodes, for reading, searching and quoting.
• SRT — numbered cues with start and end times, the format most players and platforms accept.
• VTT — the web-native equivalent, used by HTML5 audio and video players.

Audio formats accepted

MP3, WAV, M4A, AAC, FLAC, ALAC, OGG, Opus, WMA, AC3, DTS and MKA, up to 2 GB per file. Compressed formats work, but a heavily compressed MP3 loses detail the recogniser needs — if you have the original WAV or FLAC, use that instead of a re-encoded copy.

What people transcribe

• Podcast episodes, for show notes, search and repurposing into written posts.
• Interviews and research calls, where a searchable record matters more than a subtitle track.
• Lecture and seminar recordings, turned into revision notes.
• Voice memos and dictated drafts.
• Radio and archive material being catalogued.