Speech-to-Text (STT)
Last updated June 10, 2026
What is Speech-to-Text in simple terms?
In simple terms, speech-to-text turns talking into typing. It listens to spoken words and writes them out as text — the technology behind voice dictation, live captions, and the transcripts of voice messages.
What is Speech-to-Text?
Speech-to-text (STT) is technology that converts spoken language into written text, automatically transcribing what someone says into words on a screen.
Speech-to-text (STT) does exactly what its name says: it takes spoken audio and turns it into written words. Also called automatic speech recognition, it's the technology behind dictating a message instead of typing it, the live captions that appear under a video, the transcript of a voice memo, and the first step in any voice assistant — before a system can act on what you said, it has to convert your speech into text it can work with. It's one of the most quietly widespread AI capabilities, running in the background of countless everyday interactions where talking is easier or more accessible than typing.
Getting this right is harder than it sounds, which is why it took AI to do it well. Human speech is enormously varied: people have different accents, speak at different speeds, slur and mumble, talk over background noise, and use words the system has never encountered, from names to slang to technical jargon. Older speech-recognition systems, built on hand-crafted rules, were rigid and easily defeated by all this variation. The leap in quality came from deep learning — training models on vast amounts of recorded speech paired with accurate transcripts, so the system learns the messy, flexible relationship between sounds and words by example rather than by rule. That's what made today's speech-to-text reliable enough to depend on in real situations rather than just controlled ones.
Speech-to-text is the natural counterpart to text-to-speech, which goes the other way and turns written words into spoken audio; together they form the bridge between how people prefer to communicate and how machines handle information. STT also matters a great deal for accessibility, opening up technology to people who can't easily type and providing captions for those who are deaf or hard of hearing. It still has real limitations — accuracy drops with heavy background noise, strong or unfamiliar accents, overlapping speakers, and specialized vocabulary — and like any system that learns from data, it can perform unevenly across different groups of speakers if its training data underrepresented them. But it has become accurate and fast enough that talking to your devices, and having them faithfully write down what you said, is now an ordinary part of daily life.
Real-world example of Speech-to-Text
Two colleagues join a video call, and one of them is hard of hearing. As the meeting goes on, live captions scroll across the bottom of the screen, turning everything the other participants say into text in close to real time, so she can follow the conversation by reading along. When someone speaks quickly or there's a burst of background noise, the captions occasionally fumble a word — but for the most part they keep pace accurately enough that she takes part fully, no different from anyone else on the call. That seamless, instant conversion of speech into readable text, making a meeting accessible to someone who couldn't otherwise hear it, is speech-to-text doing one of its most valuable jobs.
Related terms
Frequently asked questions about Speech-to-Text
What is the difference between speech-to-text and text-to-speech?
They're opposites that work as a pair. Speech-to-text takes spoken audio and converts it into written words — it listens and transcribes. Text-to-speech does the reverse, taking written text and converting it into spoken audio — it reads aloud. So one turns talking into typing and the other turns typing into talking. Voice assistants use both: speech-to-text to understand what you said, and text-to-speech to reply out loud. They're complementary halves of letting people and machines communicate by voice.
How does speech-to-text work?
Modern speech-to-text uses models trained with deep learning on enormous amounts of recorded speech paired with accurate written transcripts. By processing all those examples, the system learns the complicated, flexible mapping between the sounds of speech and the words they represent — across different voices, accents, and speaking styles — rather than relying on rigid hand-written rules. Once trained, it can take new audio it has never heard and produce a written transcription, handling much of the natural variation in how real people speak.
What is speech-to-text used for?
A wide range of everyday tasks: dictating messages, notes, and documents instead of typing; live captioning of videos, meetings, and broadcasts; transcribing voice memos, interviews, and calls; and serving as the first step in voice assistants, which must convert speech to text before acting on it. It's especially important for accessibility, helping people who can't easily type and providing captions for those who are deaf or hard of hearing. Accuracy can still drop with heavy noise, strong accents, or specialized vocabulary.