Stockertim Dictionary · Artificial intelligence · Technology
Speech-to-text
Also called STT, automatic transcription, voice-to-text
In short
Speech-to-text is technology that converts spoken audio into written text, such as a meeting transcript, and can mishear names, numbers and who was speaking.
- Beginner
- Technology
- Checked
In detail
Speech-to-text software listens to recorded or live audio and produces written words. Modern systems use neural networks trained on large amounts of transcribed speech: the audio is broken into short slices, the model estimates which sounds and words are most likely, and a language model helps choose between similar-sounding options based on context.
Many tools add extra steps. Speaker labelling tries to work out who said each line. Punctuation and paragraphs are inserted automatically. Some products pass the transcript to an AI assistant that writes a summary or a list of action items.
Accuracy depends on the recording. Clear audio from one speaker in a quiet room transcribes well; cross-talk, accents the system has heard less often, poor microphones, technical terms and personal names cause errors. Numbers are a particular weak point: fifteen and fifty, or a date said quickly, are easily confused. Speaker labels can be wrong, which matters when a transcript records who agreed to what.
Using speech-to-text at work also raises consent and confidentiality questions. People in a meeting should be told before it is recorded or transcribed, the transcript is a record that may contain personal data, and where the audio is processed and how long it is kept depend on the tool's settings and the organisation's policy.
Worked example
A project lead at Lagoon Logistics checks an automatic transcript of a one-hour planning meeting before sharing it.
- She compares a 1,500-word section against the recording and finds 45 wrong words: 45 / 1,500 = 3%
- Of the 45 errors, 9 are names, figures or dates: 9 / 45 = 20%
- She corrects those 9 first, because they change meaning, then fixes two wrong speaker labels.
The shared transcript has every name, figure and speaker label checked against the recording.
In practice
Speech-to-text is used for meeting notes, interview transcripts, captions, dictation and call-centre records. Common practice is to announce recording at the start, correct names and numbers before sharing, and follow the organisation's rules on where transcripts are stored.
Watch out
- A transcript is not a verbatim record until someone has checked it; speaker labels and numbers are the usual weak points.
- Speech recognition is the wider field; speech-to-text is its most familiar workplace use.
Learn it properly
Official sources
- Artificial intelligence · Information Commissioner's Office
- Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile · National Institute of Standards and Technology
Facts last checked on 3 November 2026.
Education, not advice: the Stockertim Dictionary explains what terms mean. It never recommends buying or selling anything.