Glossary
What is speech to text?
By the KeyMo team · Updated
Speech to text is software that listens to spoken audio and writes it out as text. It is the first step behind automatic captions and transcripts. The output is a draft: it can mishear names and small words, so anything people will read should be checked.
Speech to text, defined
Speech to text, also called speech recognition or dictation, takes a recording or a live voice and produces words. A speech to text app, tool or software package may write plain text, or text with a time for each word. The 'AI' in speech to text ai is shorthand for machine learning, and YouTube describes its own automatic captions as coming from machine learning.
In schools and workplaces, 'speech to text' can also name an assistive tool that types what a person dictates. That is a use of the same technology, not a different technology.
What it does and does not do
| It usually does | It usually does not |
|---|---|
| Turn clear speech into words | Know names, brands and jargon it was never shown |
| Add a time to each word or line | Understand meaning, so a missed 'not' flips the sentence |
| Make a first transcript quickly | Decide who is speaking, unless it has diarization |
| Feed captions and search | Catch jokes, sarcasm or crosstalk reliably |
The W3C gives a sharp example: miss one word such as 'not' and the caption can contradict what was said. Its position is that automatic captions do not meet accessibility needs unless confirmed fully accurate, and that they usually need significant editing. YouTube says the same in plainer terms: quality varies, and automatic captions should always be reviewed and edited.
From audio to captions on a clip
- Speech to text turns the audio into words with timing.
- The words are grouped into lines or shown word by word.
- A caption style adds font, colour and motion.
- A person checks the result on the finished clip.
Step 4 is the one skipped most. A transcript hiding in a text file can be wrong without anyone noticing. A caption burned into a clip that thousands of people watch cannot.
Where KeyMo touches it
KeyMo's clips come with captions in 12 styles, shown word by word, or no captions at all. We do not quote an accuracy figure, and you should treat the words as a first draft. Watch each clip before you post it. Related terms: transcription, transcript and word-by-word captions.
Common mistakes
- Trusting the first pass because it looks tidy.
- Using it on bad audio and expecting a clean result. YouTube's Auto-sync tool is not recommended for videos over an hour or with poor audio.
- Confusing it with text to speech, which goes the other way.
- Publishing a transcript with 'um' and 'ah' left in when the goal was readable captions.
Sources: W3C WAI: Captions/Subtitles (checked 11 October 2026); YouTube Help: Use automatic captioning (checked 11 October 2026); YouTube Help: Add subtitles and captions (checked 11 October 2026).