Turning speech into text

An hour-long meeting, a voice memo, a podcast — and seconds later every word is written out, timestamped and searchable. How does audio become text you can read and search?

The reverse trick

The last lessons went from text to speech. This one runs the other way: speech to text. Feed the tool an audio recording — a meeting, an interview, a lecture — and it writes down what was said, word for word, with a timestamp on each line so you can jump straight to any moment.

Audio in, text out

It doesn't just dump a wall of words. Good transcription tags each line with the time it was spoken, often labels who's speaking, and lets you search the whole recording for a word — so a two-hour call becomes something you can skim in a minute. Press Play below and watch the words appear in step with the audio, each stamped with its moment.

Where it earns its keep

This quietly saves hours: automatic subtitles and captions, meeting notes you didn't have to type, searchable interviews for a journalist, a doctor's dictated notes. Tools like OpenAI's Whisper, Deepgram and AssemblyAI do the heavy lifting. It isn't perfect — names, jargon, crosstalk and strong accents still cause slips — so skim the result before you trust it.

The gist

• —Transcription (speech-to-text) turns spoken audio into written words — the reverse of text-to-speech
• —It timestamps each line and often labels speakers, so a long recording becomes searchable and skimmable
• —Powers subtitles, meeting notes and searchable interviews (Whisper, Deepgram, AssemblyAI); check it for slips on names, jargon and accents